Download The Theory of Polynomial Functors

THE
THEORY
POLYNOMIAL
QIMH
OF
FUNCTORS
XANTCHA
Doctoral Dissertation
Department of Mathematics
Stockholm University
c Qimh Xantcha 2010
isbn 978-91-7447-190-8
Printed by us-ab
3
Jag vill icke säga Licentiatens namn, men initial-bokstafven var X.
Carl Jonas Love Almqvist, Svensk Rättstafnings-Lära
4
ACKNOWLEDGEMENTS
Mon génie étonné tremble devant le sien.
Jean Racine, Brittanicus
The epigraph contains the judgement passed by Nero on his mother Agrippina. Disregarding the peculiar circumstances under which it was uttered, the
quote, as such, is rather well suited to describe the unprecedented awe and
admiration we feel for our advisor, the venerable Professor Torsten Ekedahl.
It was a startling encounter, back in the summer of 2006, when we first
entered his oHce, coyly proclaiming ourself to be his new graduate student.
“Oh yes,” he said, peering over his spectacles, “I remember you. I mean,
I don’t remember you, but I recall your looking something like that. Now,
what should you start with? Considering you know so little algebra1 , I was
thinking you could begin with a little starter. How would you like to explore
the connection between polynomial and strict polynomial functors?”
Not knowing better, we acquiesced, mainly because the word “polynomial” did not ring any alarm bells. It thus all began like an appetiser. It ended
up a doctoral thesis.
The project grew under our hands, expanded in all directions, and we
watched with pleasure a beautiful theory taking shape. Conducting research
may be likened to exploring an unknown territory, but many a time we have
felt less like an Explorer, and more like an Architect. By the point we arrived
at the scene, the state of avairs was a miserable one, a malaria-infested swamp
of murky waters, a jungle of buzzing mosquitoes and tangled undergrowth.
But over the course of these five years, we have worked hard to clear the
ground, dike the land, and cut down the trees to erect a glorious palace at
their place, crowned with towers and turrets glistening in the sun and coloured
banners flapping gaily in the breeze, amidst cascades of hanging gardens. Our
mission has indeed been the Architect’s.
The theory, as we here present it, is beautiful; or so we feel. There remains
to be seen if it can also be useful.
1 This
defect has since been remedied, we hope.
5
6
INTRODUCTION
[. . . ] but luckily Owl kept his head and told us that the Opposite of an Introduction,
my dear Pooh, was a Contradiction; and, as he is very good at long words, I am
sure that that’s what it is.
Alan Alexander Milne, The House at Pooh Corner
Three questions concerning the subject at hand, polynomial functors, are begging to be answered. What are polynomial functors, where do they come
from, and what are they good for?
The latter two are most easily replied to. Polynomial functors (the weaker
notion) were introduced by Professors Eilenberg and Mac Lane in 1954, who
used them to study certain homology rings ([6]). Strict polynomial functors
were invented by Professors Friedlander and Suslin in 1997, in order to develop the theory of group schemes ([10]). Since then, the two spurious concepts
have evolved side by side. Mentions of them have appeared scattered in articles, generally revolving around the themes of homotopy and homology. The
stance taken is rather a pragmatic one, usually treating polynomial functors as
a means, rather than an end in themselves.
As far as we know, no cross-fertilisation has yet taken place. This treatise is
likely the first ever to actually interrelate the two species. That, we allege, is the
ultimate end of this work: a comparison of polynomial2 and strict polynomial
functors.
What, then, is a polynomial functor? Let us consider the category Z Mod
of abelian groups. Two familiar functors on this category are the (co-variant)
Hom-functor
HompP, q
and the tensor functor
Q b ;
P and Q being fixed groups. They are both additive in the following sense:
HompP, α βq pα βq α β
Q b pα βq 1Q b pα βq 1Q b α
HompP, αq HompP, βq
1Q b β Q b α Q b β;
2 Of course, as we discovered in due time, polynomial functors provide much too weak a
notion. Over more general base rings than Z, they are subsumed by numerical functors.
7
8
α and β denoting homomorphisms. Consider now the tensor square T 2 , given
by the equation
T 2 pM q M b M.
It still maps homomorphisms to homomorphisms, but it is itself not additive,
for
T 2 pα
βq pα
whereas
βq b p α
T 2 pαq
βq α b α
T 2 pβq α b α
αbβ
βbα
β b β,
β b β.
Evidently, if there be any justice in the world, this functor should belong to
the quadratic family. The question is how to formalise this.
One approach is to observe that, while T 2 does not satisfy the aHnity
relation
T 2 pα βq T 2 pαq T 2 pβq T 2 p0q 0
(T 2 p0q 0 gives additivity), it will, however, satisfy the higher-order equation
T 2 pα
β
γq T 2 pα βq T 2 pβ
T 2 pαq T 2 pβq
γq T 2 pγ αq
T 2 pγq T 2 p0q 0.
This is what it means to be quadratic in the sense of Eilenberg and Mac Lane.
But the functor T 2 not only behaves like a polynomial, Friedlander and
Suslin argued, it is in fact a polynomial. To motivate such a designation, they
overed the following calculation:
T 2 paα
bβq paα bβq b paα bβq
a2 pα b αq abpα b β β b αq
b2 pβ b βq.
One would think that, in order to discuss polynomial functors, need would
arise for things such as the “square of a homomorphism”, but not so. It will, in
fact, be suHcient that the coefficients a and b of the homomorphisms transform
as quadratic polynomials. This is what it means for the functor to be strict
quadratic.
Additive functors have been extensively studied; non-additive functors less
so, and rarely for their own sake. The first real investigation of their properties was not performed until 1988, when Professor Pirashvili showed that
polynomial functors are equivalent to modules over a certain ring ([17]), a
result we shall build upon and generalise. A similar study was conducted on
strict polynomial functors in 2003 by Dr. Salomonsson, our predecessor, in his
doctoral thesis [20].
A radically diverent method of attack was initiated by Dr. Dreckman and
Professors Baues, Franjou, and Pirashvili in the year 2000. Their approach
was to combinatorially encode polynomial functors, for this purpose utilising
the category of sets and surjections. Evidently inspired by this device, Dr. Salomonsson would later repeat the feat for strict polynomial functors, employing instead the category of multi-sets.
9
Such is the theory of polynomial functors as it stands today — or rather as
it stood just recently. This thesis proposes the following:
1:o. To generalise the notion of polynomial functor to more general base
rings than Z, so that it smoothly agree with the existing definition of
strict polynomial functor, allowing for easy comparison. This results in
the definition of numerical functors (Chapter 6).
2:o. To make an extensive study of numerical maps of modules (which will
be needed so as to properly understand the functors), to see how they fit
into Professor Roby’s framework of strict polynomial maps (Chapter 5).
3:o. To conduct a survey of numerical rings (in order to understand the maps).
This has, admittedly, been done before, in a somewhat diverent guise,
but our approach will be seen to contain a few novelties (Chapter 1).
4:o. To develop the theories of numerical and strict polynomial functors
(Chapter 7) so that they run (almost) in parallel (Chapter 8).
5:o. To show how also numerical functors may be interpreted as modules
over a certain ring (Chapter 9).
6:o. To expound the theory of mazes (Chapter 3), which will be seen to vastly
generalise the category of surjections employed by Professor Pirashvili
et al., since they turn out to encode, not only polynomial or numerical
functors, but all3 module functors over any4 base ring (Chapter 10).
7:o. To simplify Dr. Salomonsson’s construction involving multi-sets (Chapter 2), making it more amenable to a comparison with mazes (Chapter
4).
8:o. To prove comparison theorems interrelating numerical and strict polynomial functors (Chapter 11).
9:o. And, finally, to merely indicate (Chapter 12) how polynomial functors
may be used to extend the operad concept, a line of thought already
present in Dr. Salomonsson’s thesis.
3 Fine
4 Fine
print: right-exact and commuting with inductive limits.
print: unital.
10
POLYNOMIELLE
PÅ
FUNCTORER
MODULE-CATEGORIER
Sedo-Lärande
Tankor
öfver
Algebran
Wi hade litet at fäya, om Wi allenaft bewifte at twå och twå äro fyra.
Olof Dalin, Then Swänska Argus, 1732 N:o 1
Sedan tidernas begynnelfe har mennifkjan egnat fig åt Arithmetik, hvarmed oftaft plägar menas manipulationer af tal medelft de fyra räkne-fätten addition,
fubtracion, multiplication och divifion. Detta är hvad Mathematik de flefta
mennifkjor någonfin komma i contac med, emedan det är, dels hvad de få lära
fig i fcholan, och dels hvad fom någorlunda eger tillämpning i hvardagslifvet.
Men med en dylik begränfad upfattning om Mathematikens väfen torde
man förundra fig ftorliga deröfver, at det alls bedrifves forskning inom Mathematik. Känner man icke redan allt om de fyra räkne-fätten, frågar fig den
mindre kunnige, och kunna des utom icke våra moderna räkne-machiner utföra defsa operationer långt qvickare än någon menfklig hjerna?
Det är vifserligen helt rigtigt, det man icke forfkar inom Arithmetik. At
correc utföra enkla räkne-operationer har mennifkjan kunnat fedan urminnes
tider. Strengt taget anfer man egentligen ej Arithmetican, för all fin tillämplighet, vara Mathematik per fe; fnarare betragtas hon få fom någon form af
Ämne
Arithmetik
Elementar Algebra
Abftrac Algebra
Categorie-Theorie
(“Abftrac Non-Sens”)
?
Uptäckt
Tidernas
begynnelfe
1500-talet
1800-talet
Exempel på Objecter
Tal: 0, 1, 7, 31 , 2, π, i, . . .
1900-talet
Categorier: Grp, Rng, Fld, X
Vec, Mod, . . .
Variabler: x, y, z, . . .
Algebraifka ftrucurer:
Grouper, ringar, kroppar,
lineara rum, moduler, . . .
Exempel på Equationer
p1 2q 4 1 p2 4q
px yq z x py zq
px yq z x py zq
bX bX
b
1 µ
X
µ
bX
Tabell 1: Hiftorique öfver Algebrans Utveckling.
11
b /
X bX
µ 1
µ
/
X
12
räkne-lära. Hon tilhandahåller endaft de fimplafte af verktyg, hvilket bevifas
deraf, at par exemple de gamle Ægyptierna, trots 3,000 år af obruten civilifation, icke voro capable at löfa den quadratifka equationen.
Detta förefaller måhända den moderne läfaren en fmula befynnerligt, ty at
de tvenne rötterna til equationen
x2
px
gifvas af
p
x
2
q0
c p 2
2
q,
det torde hvar och en erinra fig från fin fchol-tid. Men för Ægyptierna gick
denna formul fynbarligen ouptäckt i 3,000 år. Raifonen härtil är icke fvår at
förftå: utan tilgång til formler, voro de hänvifade til långa och omftändliga
befkrifningar i ord, för at nedteckna allmänna reglor för equationers löfande.
Läfaren kan fjelf förföka fig på, at befkrifva formulan ofvan i ord. Af den
concentrerade formeln lär blifva en blafkig nouvelle; och af des härledning —
en hel roman! Intet under, det Ægyptierna aldrig funno den!
Ej heller de gamle Indierna kunde, för få vidt man vet, löfa den quadratifka
equationen. At det lyckades Babylonierna och Chineferna, trots at äfven defse
voro hänvifade til befkrifningar i ord, får fes fom en fmärre bedrift.
Så pafserade den ftora Revolutionen. Remarquabelt nog, timade denna
famtidigt med öfriga culturella omhvälfningar af famhället, det vil fäga, under
Renaifsancen, då man uptäckte den fymbolifka algebran. Man fant alltfå på
konften at fkrifva formler. Och utan formler, ingen Mathematik — då återftår
endaft räkne-lära. Det är fåledes fvårt at öfverfkatta den fymbolifka algebrans
betydelfe för Mathematiken, och man fkulle kunna likna des införande vid
hjulets upfinnande.
Den Mathematifka Wetenfkapens arbetsfält vidgades med ens, och öppnade up för hvad fkulle kunna benämnas den Elementara Algebrans epoque. Det
nya formel-fpråket gjorde omedelbar fuccès, och framgången lät icke vänta på
fig. Inom kort lyckades det Italienfka Mathematiker at löfa fåväl den cubifka,
fom ock den quartifka, equationen.
Den Elementara Algebran använder fig af variabler i ftället för tal, och marquerar öfvergången från ziver-räkning til bokftafs-räkning. Des ftyrka ligger
ej blott deri, at den möjliggör nedfkrifvandet af equationer affedda at lösas;
hon låter ofs ock formulera allmängiltiga räkne-lagar, fanna för alla tal.
Det förhållande par exemple, at det, då tvenne tal fkola adderas, helt faknar betydelfe i hvilken ordning defsa tagas, kan då compac och öfverfkådligt
fkrifvas fom den Commutativa Lagen
x
yy
x,
giltig för alla tal x och y.
Det förhållande åter, at det, då man har at addera trenne tal (i någon gifven
ordning), icke fpelar någon rôle i hvilken ordning de tvenne additionerna ut-
13
föras, utgör den Associativa Lagen
px
yq
zx
py
zq.
Mathematiker egna fig fåledes, tvert emot den föreftällning folk i gemen
hafva om defse, i påfallande liten utfträckning åt zivror. (Jemför det vulgaira
uttrycket “ziverkarl”!)
Näfta abftracion egde rum under 1800-talet, då man obferverade, at flere
af de lagar, hvilka gälla för räkning med vanliga tal, äfven ega giltighet i andra
fammanhang. Man noterade exempelvis, at den Afsociativa Lagen för addition:
px
yq
zx
py
zq,
finner fin motfvarighet för multiplication:
px yq z x py zq.
Det exifterar vifserligen ockfå en afsociativ lag för addition af vecorer:
pu
vq
wu
pv
wq,
multiplication af matricer:
pA Bq C A pB C q,
famt för compofition af funcioner:
pf gq h f pg hq.
Som fynes kunna räkne-operationerna vara af de meft fkilda flag, och de ingående ftorheterna äro ej längre begränfade til at vara tal. Det interefsanta är
fåledes ej längre räkne-operationena fom fådana, långt mindre hvilka objecer de
verka på, utan operationernas gemenfamma egenfkap associativitet.
Man infåg, det vore af ftörfta betydelfe, at ifolera detta phenomen, och tog
fig före, at ftudera algebraiska structurer. Defsa äro mängder af Mathematifka
objecer, equiperade med en eller flera operationer, hvilka må upfylla vifsa
axiomata.
Sålunda är exempelvis en half-groupe en mängd med en enda operation ,
fatiffierande juft den afsociativa lagen
px yq z x py zq
(för alla element x, y, z tilhörande denna mängd). Hvad operationen “betyder” är icke längre relevant; det kan vara addition, multiplication, compofition; men ockfå en helt abftrac operation, utan förankring i hvardagslifvet.
Det interefsanta är egenfkapen denna poftuleras befitta: afsociativitet.
Man hade nu abftraherat up ännu en niveau och entrerat den Abstracta Algebrans domaine. En af den Abftraca Algebrans tidigafte triumpher var hennes bevis för den quintifka equationens olöflighet medelft clafsifka algebraifka verktyg (få benämnde radicaler; defse äro de fyra räknefätten jemte rotutdragningar), få at ingen traditionel löfningsformel exifterar i detta fall.
14
Studerades fålunda ej längre “en räkne-operation i taget”, utan “alla famtidigt”. Vi marquerade detta ofvan genom at beteckna den godtyckliga (okända)
operationen med en ftjerna , precis fom man tidigare hade betecknat tal med
bokftäfver, för at ftudera, icke “et tal i taget”, utan “dem alla famtidigt”. Detta
kan fägas vara Mathematikens väfen: fökandet efter univerfella, allmängiltiga
principer.
Et fynfätt, fom introducerades vid denna tid, och fedan des vunnit laga
häfd, är det, at det icke få mycket är de algebraifka ftrucurerna fjelfva, hvilka
förtjena et ftudium, fom afbildningarna dem emellan. Defsa afbildningar må
icke vara af helt godtyckligt flag, utan dem vare det ålagdt, at bevara de ingående räkne-operationerna. En fådan ftrucure-bevarande afbildning kallas
homomorphism. Hafva vi exempelvis en enda operation är criterium, för at
afbildningen ϕ vare en homomorphifm, det, at
ϕpx yq ϕpxq ϕpyq.
Det är fåledes detta, der ftuderas inom den Abftraca Algebran: algebraifka
ftrucurer, tilfamman med deras homomorphifmer.
Algebraifka ftrucurer äro legio. Förutom half-grouper hafva vi, för at
blott nämna de meft frequent förekommande: grouper, ringar, kroppar, lineara rum, moduler, . . . . Hvar och en af defse clafser characeriferas af et antal
operationer med tilhörande axiomata, och hvar och en egnas et ingående ftudium inom correfponderande Mathematifka difcipline: Groupe-Theorie, RingTheorie, Kropp-Theorie, Linear Algebra, Module-Theorie, . . . .
Under 1900-talet begynte Mathematikerna fyftematifera i denna brokiga
flora (eller fauna?) af algebraifka ftrucurer. Likafom deras föregångare hundra
år tidigare hade obferverat likheter mellan räkne-lagarne för olika operationer,
noterade man nu vifsa reglor och lagbundenheter, en vifs Method in the Madness,
alla defsa algebraifka ftrucurer emellan. Begrepp rörande en vifs forts ftrucure befunnos ega motfvarigheter för en annan; vifsa theoremer för en theorie
beviftes vara fanna äfven inom andra theorier; etc. Önfkan väcktes, at ena denna rika famille af algebraifka ftrucurer under en och famma parapluie.
Vid förfta revolutionen hade man abftraherat bort talen, i det defsa erfattes af variabler. Under andra refan abftraherades fedan vederbörligen räkneoperationerna bort, at remplaceras med godtyckliga fådana. I focus placerades
nu ftrucurerna defsa operationer gåfvo uphof til. Under tredje vågen i Algebrans utveckling, abftraherade man flutligen bort äfven ftrucurerna. Man ville
ftudera, icke en fpecifique algebraifk ftrucure — precis fom man tidigare valt,
at icke ftudera et fpecifique tal eller fpecifique räkne-operation — utan dem
alla famtidigt.
Studium inleddes af få kallade categorier, hvarmed betecknas famlingen af
alla algebraifka ftrucurer af et gifvet flag. Hafva vi fåledes: categorien af grouper, categorien af ringar, categorien af kroppar, categorien af lineara rum, categorien af moduler, . . . Man fökte utröna, dels det inre machineriet i defsa
categorier, dels deras relationer fins emellan (hvilken fträfvan kan anfes fammanfatta en Algebraikers lifsnäring: at utforfka structure).
15
Men likfom et ftudium af algebraifka ftrucurer vore otänkbart utan et
famtidigt ftudium af deras homomorphifmer, det är: ftrucure-bevarande afbildningar; är det omöjligt at tänka fig categorier utan ftrucure-bevarande
afbildningar dem emellan. Defsa gå här under namnet functorer.
Som exempel anföre vi categorien af moduler. En funcor på denna categorie tager en module och producerar en annan; famtidigt fom hvarje homomorphifm transformeras til en annan homomorphifm. Detta må då icke fke
helt godtyckligt, utan åter fkola vifsa axiomata upfyllas.
Man fant ftraxt, at det för fomliga categorier vore meningsfullt, at tala om
lineare funcorer. Användningsområden för defse äro mångfaldiga, och deras
theorie följagteligen mycket väl utforfkad. Men juft categorien af moduler har
des utom vifat fig befitta en liten egenhet, hvars rätta betydelfe man egentligen förft på fednare tid infett. Det är nämligen möjligt för funcorer på denna
categorie at vara (i någon mening) polynomielle, det vil fäga: quadratiske, cubiske, etc. Detta är themat för föreliggande afhandling: Polynomielle Functorer på
Module-Categorier.
Huru fkal då begreppet “polynomialitet” lämpligen definieras? Vi lägge
fram tvenne möjliga definitioner: det är begreppen numerisk funcor, refpecive strict polynomiel funcor. Defsa te fig vid förfta anblicken väfensfkilda, men
vi bevife, det de i fjelfva verket äro nära beflägtade. Man finner, at en ftric
polynomiel funcor par nécefsité måfte vara numerifk, medan det omvända
icke gäller — det ena begreppet är fåledes fvagare än det andra. Men huru
mycket fvagare? Under hvilka omftändigheter är en numerifk funcor ftric
polynomiel? Kan en numerifk funcor på något lämpligt vis approximeras af
en ftric polynomiel fådan? Det är frågor fom defsa vi dryfta, och de utgöra
goda exempel på hvilken forts fpörsmål Mathematiker egna fig åt.
Det categorifka tänkandet genomfyrar i dag ftora delar af Mathematiken.
Vi tacke då Categorie-Theorien icke få mycket för de theoremer hon bragt ofs,
utan för hennes philosophie. Hon har fkänkt ofs et communt fpråk för Mathematiken, eller i hvart fall de delar, hvilke äro någorlunda beflägtade med
Algebran. Det måfte nu påpekas, det icke alla Mathematiker prifa CategorieTheoriens förtjenfter fans réfervation. Man har myntat termen abstract non-sens
för denna gren af Mathematiken, hvars theoremer ega fådan allmängiltighet,
at de icke fäga någon ting alls.5
Categorie-Theorie kan fägas utgöra den ultimata abftracionen — ultimata,
emedan det exifterar et få finguliert objec fom Categorien af Categorier. Men
Categorie-Theorien är nu ökänd juft för fin förmåga, at flå knut på fig fjelf.
5 Abftracion kan löpa amok. Enligt upgift hafva äfven vifse delar af Univerfel Algebra lidit af
detta. Det blef tommare och tommare.
16
Chapter 0
PRELIMINARIES
En tyktes wara hwaß, och grep mig an för stöld,
At jag ur böcker tog, med andras tankar jäste;
Men huru wet hon det, som aldrig nånsin läste?
Hedwig Charlotta Nordenflycht,
Satyr emot afwundsjuka Fruentimber
§1. Set Theory
We will use standard notation for finite sets:
rns t1, . . . , nu.
The text is pervaded by the use of multi-sets, which are given a detailed
introduction in Chapter 2.
§2. Module Theory
It is not uncommon for algebraists to insist upon the existence of a multiplicative identity in each abstract ring. Where this inclination stems from is
diHcult to say, but the end result is that ideals are begrudged the right to be
rings. We shall not follow this trend, but magnanimously allow any abelian
group endowed with an associative, bilinear multiplication to call itself ring.
That being said, we point out that all our rings will indeed possess an identity,
but we shall always prefix them with the word “unital”.
Since all rings under consideration will possess an identity, we shall tacitly
assume all modules to be unital. Moreover, modules will be left modules,
unless explicitly said otherwise.
The theory of polynomial maps and functors will be developed over a fixed
base ring B. Consequently, we adopt the following conventions.
I. All modules will be unital left B-modules (unless otherwise stated), and
Mod will denote the category of these.
17
18
Chapter 0. Preliminaries
II. All algebras will be commutative and unital B-algebras (we shall only
have reason to consider algebras in the case when B is commutative),
and CAlg will denote the category of these.
III. When B is numerical, all numerical algebras will be assumed B-algebras,
and NAlg will denote the category of such.
IV. Linearity will always mean B-linearity, and all homomorphisms will be Bmodule homomorphisms. (However, we will consider general maps of
modules, and they will usually be very non-linear!)
V. Tensor products will be computed over B, unless otherwise stated.
VI. A linear category denotes a B-linear category; by which is meant a category enriched over Mod, so that its arrow sets are in fact B-modules.
(In the case B Z, this is what is known as a pre-additive category.)
VII. The following additional assumptions will be placed upon the base ring:
(a) When discussing arbitrary maps and functors, B can be an arbitrary
ring (in particular, it may possibly be non-commutative).
(b) When discussing numerical maps and functors, B will of course be
assumed numerical.
(c) When discussing strict polynomial maps and functors, B will be assumed commutative only.
At his1 leisure, the reader may put B Z, and everywhere substitute “abelian group” for “module”.
§3. Category Theory
Now for some general notation concerning categories. When C is a category,
C
will denote the opposite category. The set of arrows from an object X of C to
another object Y will be denoted by
C pX, Y q.
There are three exceptions to this rule. Inside a module category, the set
of homomorphisms between the R-modules M and N will be denoted by
HomR pM, N q,
1 Of course, women, with their greater capability of multi-tasking, will no doubt be able to
keep in mind the more general case.
§4. Semi-Abelian Category Theory
19
and the letter R will be omitted if the ring is clear from the context (which it
usually is). The set of endomorphisms of a module M will be denoted by the
symbol
End M.
Furthermore, in the category of categories, we will let
FunpA, Bq
denote the category of functors from A to B; sometimes linear, and sometimes
not, depending on context. And finally, inside a functor category, the natural
transformations between two functors F and G will be denoted by
NatpF, Gq,
which will be shortened to
Nat F
in the case F G.
The following is a (not exhaustive) list of the categories we will use. Those
which are not standard will be defined in the text.
Set Sets.
MSet Multi-sets.
Laby The labyrinth category.
Sur Sets with surjections.
CRing Commutative, unital rings.
CAlg Commutative, unital algebras.
NRing Numerical rings.
NAlg Numerical algebras.
Mod Modules.
FMod Free modules.
XMod Finitely generated, free modules.
Num Numerical functors.
QHom Quasi-homogeneous functors.
SPol Strict polynomial functors.
Hom Homogeneous functors.
§4. Semi-Abelian Category Theory
The proposition below is (un)known to mathematicians as Delsarte’s Lemma,
but there seems to be no tangible way to attach Professor Delsarte’s name
20
Chapter 0. Preliminaries
unto it.2 In our licenciate thesis, we gave two versions, with rather divering
proofs, one for rings and one for abelian categories. We much deplored this,
being firm adherents to Professor Bourbaki’s infamous slogan never to prove
a theorem that could be deduced as a special case of a more general theorem.
It is not without some pride that we now enunciate the following ultimate
Delsarte’s Lemma. For an introduction to semi-abelian categories, we refer to
Professor Borceux’s survey paper [2].
Theorem 1: Delsarte’s Lemma.
• Let A, B, and C be objects of a semi-abelian category such that C „ A B and the
projections C Ñ A and C Ñ B are regularly epic. Then A and B have a common
quotient object
A
a
//Doo
b
B,
which completes the square into a Doolittle diagram3 :
<AE
yy< O EEE a
y
EE
yy p
EE
yy
E" "
y
y c /
AB
D
CE
EE
y< <
EE q
yy
y
EE
y
EE" yyy b
" y
B
• Conversely, let a common quotient object
A
a
//Doo
b
B
of A and B be given, and let C be the pullback. Then C is a subobject of A B,
and the projections on A and B are regularly epic.
Proof. We are grateful to Professor Ekedahl and Dr. Bergh for furnishing us
with the proof, which overs an excellent opportunity to see all the classical
isomorphism theorems in action. By the Fundamental Homomorphism Theorem,
A C { Ker pc,
B C { Ker qc;
and so
D C {pKer pc Y Ker qcq
is a common quotient object of A and B, where
of subobjects.
Y denotes join in the lattice
2 Legend has it that the attribution was made by Professor Serre during his Collège de France
lectures.
3 Following the terminology of Professor Freyd, a Doolittle diagram is a square which is both a
pullback and a pushout square. Professor Popescu more prosaically calls them exact squares.
§4. Semi-Abelian Category Theory
21
It must be shown this yields indeed a Doolittle diagram. Denote
P
Ker pc,
Q Ker qc;
and consider the following diagram, whose rows by the Tower Isomorphism
Theorem are exact:
0
/ P {pP X Qq
/ C {pP X Qq
/ C {P
/0
0
/ pP Y Qq{Q
/ C {Q
/ C {pP Y Qq
/0
By the Diamond Isomorphism Theorem, the left vertical arrow is an isomorphism. Write
K P {pP X Qq pP Y Qq{Q,
and consider the diagram:
K
/ C {pP X Qq
/ C {Q
0
/ C {P
/ C {pP Y Qq
Because a diagram of the form:
Ker x
/X
0
/Y
x
is always a pullback square, the left square and the outer rectangle above are
both pullback squares. It now follows, from the “unusual cancellation property of pullbacks”, Theorem 2.7 of [2], that the right square is a pullback.
The pushout property is rather more trivial. Suppose the following diagram commutes:
/ C {P
C
C {Q
/T
The arrow C Ñ T is zero on P and Q, and therefore on P Y Q, which produces
a unique factorisation
C {pP Y Qq Ñ T .
Hence C {pP Y Qq is the pushout.
Let us finally turn to the converse of the theorem. It is clear that C, defined
as the equaliser of ap and bq, is a subobject of A B. Professor Freyd’s Pullback
22
Chapter 0. Preliminaries
Theorem for abelian categories (Theorem 2.54 of [9]) states that pullbacks
of (regular) epimorphisms are (regular) epimorphisms, and this proposition
remains valid in a semi-abelian (or exact) category.
Because the square is a Doolittle diagram, we have in an abelian category
the following exact sequence:
0
/ A`B
/C
/D
/0
We may then simply choose
D Coker pC
Ñ A ` Bq .
(In the general case, C may not be a normal subobject.)
§5. Abelian Category Theory
Let us say some words on abstract tensor products. Preparing for what will
eventually follow, let B be an arbitrary ring of scalars. All categories and all
functors of this section are assumed B-linear, and all modules are B-modules.
The theory briefly accounted for below can be found in Professor Popescu’s
treatise [18] on abelian categories, section 3.6. (He gives the case B Z, but
the extension to an arbitrary base ring is immediate.)
Let A be a category. We recall the classical Yoneda embedding
ΥA : A Ñ FunpA , Modq
X ÞÑ Ap, X q.
Suppose now that A is small4 , and let B be an abelian category with direct
sums. Fix a functor Q : A Ñ B. According to Theorem 3.6.3 of [18], there is a
unique functor
bA Q : FunpA , Modq Ñ B
having a right adjoint, and making the following diagram commute:
A LL / FunpA , Modq
LL
LL
LL
bA Q
Q LLL
L% B
ΥA
This we call the tensor product with Q over A. Its adjoint is the mediated
Hom-functor
BpQ, q : B Ñ FunpA , Modq
Y ÞÑ BpQpq, Y q.
4 When
A is finite, it is spoken of in reverence as a ring with several objects.
§5. Abelian Category Theory
Note that, when X
23
P A, the diagram above implies that
Ap, X q bA Q QpX q,
and we may then extend by right-exactness and commutation with direct
sums.
It will sometimes be convenient to have available also a reverse tensor
product. Hence, if Q : A Ñ B is a fixed functor, there is a unique functor
Q bA : FunpA, Modq Ñ B
having a right adjoint, and making the following diagram commute:
ΥA
A LL / FunpA, Modq
LLL
LLL
Qb A Q LLL
L% B
It will then be seen that, for functors Q : A Ñ B and M : A
products
Q b A M M b A Q
Ñ Mod, the tensor
are equal.
Example 1.
The special case most frequently encountered is when A tu
is a category with a single object, with End R an algebra. Then a functor
M : A Ñ Mod is a B-R-bimodule, and a functor Q : A Ñ B is a left R-object of
B. The tensor product is usually denoted by
bR Q : B ModR Ñ B,
and is uniquely specified by the equation
R bR Q Qpq
and the extension property.
Specialising even further, we may consider the case when also B
Then Qpq is simply a left R-module. Since
Mod.
R bR Q R Qpq RR bR Qpq,
the tensor product “of functors” coincides with the usual tensor product “of
modules”.
4
In some situations, there is an explicit description available for the tensor
product.
24
Chapter 0. Preliminaries
Theorem 2.
Let A and B be small categories, and fix a functor
Q : A Ñ FunpB, Modq.
Then, for M : A
Ñ Mod and Y P B, the tensor product
pM bA QqpY q
is the quotient of the module
à
P
X A
M pX q b QpX qpY q
by all relations
M pX q b QpX qpY q Q x b QpαqY pyq M pαqpxq b y P F pX 1 q b QpX 1 qpY q,
for any x P M pX q, y P QpX 1 qpY q, and α : X 1
The pair
Ñ X.
p bA Q, BpQ, qq
is, as mentioned above, an adjunction. Under certain circumstances, it is actually a category equivalence. The theorem below occurs as Corollary 3.6.4 in
[18].
Let C be an abelian category with direct
Theorem 3: The Morita Equivalence.
sums, having a full subcategory P of small projective generators, with inclusion functor
J : P Ñ C. There is a category equivalence:
p q
C J,
(
FunpP , Modq
Cf
bP J
Example 2.
Morita equivalence is most frequently used in the situation of
a single projective generator. Denoting the endomorphism ring of P tu by
R, we have
FunpP , Modq B ModR ,
and the equivalence reduces to the familiar:
C
Cd
p,q
%
B ModR
bR 4
§6. Commutative Algebra
25
§6. Commutative Algebra
The subsequent (in)equality of Krull dimensions is supposedly known by
“everybody” working in commutative algebra or algebraic geometry, and consequently impossible to reference. We are grateful to Professor Ekedahl for
furnishing the proof.
Theorem 4: Chevalley’s Dimension Argument.
non-trivial, unital ring. The (in)equality
Let R be a finitely generated,
dim R{pR dim Q bZ R ¤ dim R 1
holds for all but finitely many prime numbers p.
When R is an integral domain of characteristic 0, there is in fact equality for all but
finitely many primes p.
Proof. In the case of positive characteristic n, the inequality will hold trivially,
for then
Q bZ R 0 R{pR,
except when p | n.
Consider now the case when R is an integral domain of characteristic 0.
There is an embedding ϕ : Z Ñ R, and a corresponding dominant morphism
Spec ϕ : Spec R Ñ Spec Z
of integral schemes, which is of finite type. Letting Frac P denote the fraction
field of R{P, we may define
Cn
tP P Spec Z | dimpSpec ϕq1 pPq nu
tP P Spec Z | dim R bZ Frac P nu
tppq | dim R{pR nu Y tp0q | dim R bZ Q nu.
This latter set, by Chevalley’s Constructibility Theorem5 , will contain a dense,
open set in Spec Z if n dim R dim Z. Such a set must contain p0q and ppq for
all but finitely many primes p, so for those primes,
dim Q bZ R dim R{pR dim R 1.
Now let R be an arbitrary ring of characteristic 0. For any prime ideal Q,
R{Q will be an integral domain (but not necessarily of characteristic 0!), and
so we can apply the preceding to obtain
dim Q bZ R{Q dim R{pQ
pRq ¤ dim R{Q 1,
for all but finitely many primes p. The prime ideals of Q bZ R are all of the
form Q bZ Q, where Q is a prime ideal in R. Moreover,
L
pQ bZ Rq pQ bZ Qq Q bZ R{Q.
5 This proposition appears to belong to the folklore of algebraic geometry. An explicit reference is Théorème 2.3 of [14].
26
Chapter 0. Preliminaries
It follows that
L
dim Q bZ R max dimpQ bZ Rq pQ bZ Qq
P
Q Spec R
QPmax
dim Q bZ R{Q
Spec R
QPmax
dim R{pQ pRq
Spec R
L
max pR{pRq Q dim R{pR
P
{
Q Spec R pR
for all but finitely many p, because the maxima are taken over the finitely
many minimal prime ideals only. In a similar fashion,
L
dim Q bZ R max dimpQ bZ Rq pQ bZ Qq
P
Q Spec R
QPmax
dim Q bZ R{Q
Spec R
¤ QPmax
dim R{Q 1 ¤ dim R 1.
Spec R
Chapter 1
NUMERICAL
RINGS
At the age of twenty-one he wrote a treatise upon the Binomial Theorem, which
has had a European vogue.
Sherlock Holmes’s description of Professor Moriarty;
Arthur Conan Doyle, The Final Problem
Our licentiate thesis from 2009 opened with some lavish praise on Professor
Ekedahl, who purportedly “discovered”1 numerical rings. This is only partially correct. In his article [7] from 2002, he did indeed set forth an axiomatisation of rings with binomial coeHcients, and he is possibly the first to have
done so, but, as we were informed of only recently, such rings had in fact been
studied much earlier. Indeed, already in 1969, in connection with his work on
nilpotent groups, Professor Hall ([12]) had defined a binomial ring as a commutative, unital ring which is torsion-free and closed under the “formation of
binomial coeHcients”:
r
pr n
ÞÑ rpr 1q n!
1q
.
Of course, these two approaches are radically diverent, and it is non-trivial
that they are equivalent.
It might perhaps be argued that we ought to follow the terminology initiated by Professor Hall, the original discoverer, and use the designation binomial, rather than numerical. We have chosen to deviate from his practice, partly
because we really use the axiomatic definition proposed by Professor Ekedahl,
and partly because, we feel, the word binomial ring leads to the wrong associations: polynomial rings, and such. However, since the letter N is already
reserved for the set of natural numbers, we will (in subsequent chapters) denote our base ring of scalars by B to suggest binomial (even though it need not
necessarily be so!).
This chapter proposes to explore numerical rings for their own sake. Some
of the results can no doubt be found in the literature. We cite a recent paper
[8] from 2005, written by Dr. Elliott, which in particular aims to elucidate the
connection between binomial rings and λ-rings.
1 He
used himself the word introduced, a precaution that turned out to be wise.
27
28
Chapter 1. Numerical Rings
§1. Numerical Rings
We here present, with minor modifications, Professor Ekedahl’s axioms for
numerical rings. The original axioms were rather non-explicit, stated as they
were in terms of three mysterious polynomials, the exact nature of which was
never made precise. Our definition intends to remedy this.
Definition 1.
A numerical ring is a commutative ring with unity which is
equipped with unary operations
r
ÞÑ
r
,
n
n P N;
called binomial coefficients and subject to the following axioms:
I.
b
n
II.
a
ab
n
a
m
a
n
IV.
1
n
V.
a
0
a
p
p q n
b
.
q
ņ
a
m
m 0
III.
¸
q1
ņ
qm n
qi 1
¥
a
m
k 0
¸
k
b
q1
b
.
qm
m
k
n
n
.
k
0 when n ¥ 2.
1 and
a
1
a.
The original definition also included a (non-explicit) formula for reducing
a composition pmn q of binomial coeHcients to simple ones. Surprisingly, this
formula will be a consequence of the five axioms we have listed.
It follows easily from axioms I, IV, and V that, when the functions n are
evaluated on multiples of unity, we retrieve the ordinary binomial coeHcients,
namely
m1
mpm 1q n! pm n 1q 1, m P N.
n
a
Since nn1 1, but 0n 0 unless n 0, a numerical ring has necessarily
characteristic 0.
The numerical structure on a given ring is always unique. This will be
proved presently.
§1. Numerical Rings
29
Example 1.
In any Q-algebra, binomial coeHcients may be defined by the
usual formula:
r
r pr 1q pr n 1q
.
n!
n
4
Example 2.
For any integer m, the ring Zrm1 s is numerical. Since it inherits the binomial coeHcients from Q, it is just a matter of verifying closure
under the formation of binomial coeHcients. Because
a
f
n
p 1q p fa pn 1qq
a a
f f
n!
apa f q n!fpa n pn 1qf q ,
it will suHce to prove that whenever pi | n!, but p - b, then
pi
| pa
bqpa
2bq pa
nbq.
To this end, let
n cm pm
¤ p 1,
be the base p representation of n. For fixed k and 0 ¤ d ck , the numbers
a pcm pm ck 1 pk 1 dpk iqb,
1 ¤ i ¤ pk ,
c1 p
0 ¤ ci
c0 ,
(1)
will form a set of representatives for the congruence classes modulo pk , as will
of course the numbers
cm pm
ck 1 pk
1
dpk
i,
1 ¤ i ¤ pk .
(2)
Note that if x y mod pk and j ¤ k, then pj | x iv pj | y. Hence there are at
least as many factors p among the numbers (1) as among the numbers (2). The
claim now follows.
4
Example 3.
As the special case m 1 of the preceding example, Z itself
is numerical. For this ring there is another, more direct, way of proving the
numerical axioms. Let us indicate how they may be arrived at as solutions to
problems of enumerative combinatorics.
Axiom I. We have two types of balls: round balls, square2 balls. If
we have a round balls and b square balls, in how many ways may
we choose n balls? Let p be the number of round balls chosen, and
q the number of square balls.
Axiom II. We have a chocolate box containing a rectangular a b
array of pralines, and we wish to eat n of these. In how many ways
can this be done? Suppose the pralines we choose to feast upon
2 This is in honour of Dr. Lars-Christer Böiers, an eminent teacher, who used an example
featuring round balls and square balls during his course in discrete mathematics.
30
Chapter 1. Numerical Rings
are located in m of the a rows, and let qi be the number of chosen
pralines in row number i of these m.
Axiom III. There are a mathematicians, of which m do geometry
and n algebra. Naturally, there may exist people who do both or
neither. How many distributions of skills are possible? Let k be the
number of mathematicians who do only algebra.
Axiom IV. We are the owner of a single dog. In how many ways
can we choose n of our dogs to take for a walk?
Axiom V. Snuvy the dog has a blankets. In how many ways may
he choose 0 (in the summer) or 1 (in the winter) of his blankets to
keep him warm in bed?
4
Example 4.
The set
S
tf P Qrxs | f pZq „ Zu
of numerical maps on Z is numerical. Addition and multiplication of functions
are evaluated pointwise, as are binomial coeHcients:
f
n
pxq f pxq
n
f pxqpf pxq 1q n! pf pxq n
1q
.
Seizing the opportunity, we remind the reader that any numerical map
may be written uniquely as a numerical polynomial
f pxq ¸
cn
x
,
n
cn
P Z.
Conversely, any numerical polynomial will leave Z invariant.
4
Example 5.
Being given by rational polynomials, the operations r ÞÑ nr
give continuous maps Qp Ñ Qp in the p-adic topology. It should be well
known that Z is dense in the ring Zp , and that Zp is closed in Qp . Since the
binomial coeHcients leave Z invariant, the same must be true of Zp , which is
thus numerical.
This provides an alternative proof of the fact that Zrm1 s is closed under
binomial coeHcients. For this is evidently true of the localisations
Zppq
and therefore also for
Q X Zp ,
Zrm1 s £
Zppq .
p-m
4
Example 6.
Products and tensor products of numerical rings are numerical. Also, inductive and projective limits of numerical rings are numerical. See
Dr. Elliott’s paper [8].
4
§2. Elementary Identities
31
§2. Elementary Identities
The following formulæ are valid in any numerical ring:
Theorem 1.
1.
r
n
pr n
rpr 1q n!
2. n!
r
n
r
3. n
n
1q
rpr 1q pr n
pr n
1q
r
P Z.
when r
1q.
n1
.
Proof. The map
ϕ : pR,
q Ñ p1
tRrrtss, q,
r
ÞÑ
8̧
r n
t
n
n 0
is, by axioms I and V, a group homomorphism. Therefore, when r
ϕpr q ϕp1qr
P Z,
p1 tqr ,
which expands as usual (with ordinary binomial coeHcients) by the Binomial
Theorem. This proves equation 1. (An inductive proof would also work.)
To prove equations 2 and 3, we proceed diverently. By axiom III,
r
r
n1
r
n1
r
1
r
k 0
r
n1
n1
n1
1
r
pn 1q n 1
1̧
1
0
k
n1
1
n
r
n
n
1
k
1
k
1
1
r
,
n
which reduces to equation 3.
Equation 2 will then follow inductively from equation 3.
§3. Torsion
The word torsion will always, here and elsewhere, refer to Z-torsion. In this
section we shall prove it is absent in numerical rings. This will establish the
equivalence of numerical and binomial rings, as defined by Professor Hall.
Lemma 1.
Let m be an integer. If p is prime and pl | m, but p - k, then pl
|
m
k
.
32
Chapter 1. Numerical Rings
Proof. pl divides the right-hand side of
m mk 11 ,
m
k
k
and therefore
also the left-hand side. But pl is relatively prime to k, so in fact
m
l
p | k .
Let m1 , . . . , mk be natural numbers, and put
Lemma 2.
m m1
If
n m1
2m2
is prime, then
m|
mk .
3m3
kmk
m
tmi u ,
m n, and all other mi 0.
°
Let a prime power pl | m. Because of the relation n mi i, not all mi
unless m1
Proof.
can be divisible by p, unless we are in the exceptional case
m1
mpn
given above. Say p - mj ; then
m
tmi ui
m
mj
m mj
tmi uij
is divisible by pl according to Lemma 1. The claim follows.
Lemma 3.
Consider a numerical ring R. Let r
also mn mr 0.
Proof. Follows inductively, since if nr
r
mn
m
Theorem 2.
npr m
1q
P R and m, n P N. If nr 0, then
0, then
r
m1
npm 1q
r
m1
.
Numerical rings are torsion-free.
Proof. Suppose nr 0 in R, and, without any loss of generality, that n is prime.
We calculate using the numerical axioms:
0
0
n
nr
n
ņ
m 0
r
m
¸
q1
qm n
qi 1
¥
n
q1
n
qm
§4. Uniqueness
33
ņ
m 0
¸
r
m °
° mi m
m
m ¹ n i
t mi u i i .
mi i n
For given numbers qi , we have let mi denote the number of these that are equal
to i (of course i ¥ 1 and mi ¥ 0). Conversely, whenthe numbers mi are given,
values may be distributed to the numbers qi in tmm u ways, which accounts for
i
the multinomial coeHcient above.
We claim the inner sum
is divisible by mn when m ¥ 2. For when 2 ¤
m ¤ n 1, then m | tmm u by Lemma 2; also, there must exist some 0 j n
i
¡ 0, and for this j, Lemma 1 says n | nj m . In the case m n,
n
obviously all mi 0 for i ¥ 2, and m1 n, so the inner sum equals n1 , which
is divisible by n2 mn.
We can now
employ Lemma 3 to kill all terms except m 1. But this term
is simply r1 r, which is then equal to 0.
such that mj
j
This theorem is remarkable in all its simplicity. We know of no other example of a variety of algebras, of which the axioms imply lack of torsion in a
non-trivial way; that is, without actually implying a Q-algebra structure. Not
only that, the theorem is also a most crucial result in the theory of numerical rings. Over the course of the following sections, we will deduce several
corollaries, seemingly without evort.
§4. Uniqueness
Theorem 3.
There is at most one numerical ring structure on a given ring.
Proof. We know that
n!
r
n
rpr 1q pr n
1q,
and that n! is not a zero divisor.
§5. Embedding in Q-Algebras
Theorem 4.
Every numerical ring can be embedded in a Q-algebra, where the binomial coefficients are given by the usual formula
r
n
pr n
rpr 1q n!
1q
.
Consequently, a ring is numerical iff it is binomial.
Proof. Since R is torsion-free, the map R Ñ Q bZ R is an embedding.
34
Chapter 1. Numerical Rings
§6. Iterated Binomial Coefficients
In Z, there “exists” a formula for iterated binomial coeHcients:
r m
n
mn
¸
gk
k 1
r
,
k
(3)
in the sense that there are unique integers gk making the formula valid for
every r P Z. Professor Golomb has examined these iterates in some detail, and
his paper [11] is brought to an end with the discouraging conclusion:
No simple reduction formulas have yet been found for the most general case of
n
b
a
.
We note, however, that (3) is a polynomial identity with rational coeHcients, which means it holds in any Q-algebra, and therefore in any numerical
ring. This proves the redundancy of Professor Ekedahl’s original sixth axiom:
Theorem 5.
The formula
r m
n
mn
¸
gk
k 1
r
k
for iterated binomial coefficients is valid in every numerical ring.
§7. Homomorphisms
Let R and S be numerical rings. The ring homomorphism
Definition 2.
ϕ : R Ñ S is said to be numerical if it preserves binomial coeHcients:
ϕ
r
n
ϕpr q
.
n
S is then a numerical algebra over R.
We denote by
NRing
the category of numerical rings, and by
R NAlg,
or simply
NAlg,
the category of numerical algebras over some fixed numerical base ring R.
Theorem 6.
Every ring homomorphism of numerical rings is numerical, so that
NRing is a full subcategory of CRing.
§8. Free Numerical Rings
35
Proof. Let R and S be numerical rings, and let ϕ : R Ñ S be a ring homomorphism. Because of the absense of torsion, the equation
r
n
n!ϕ
ϕ
r
n
n!
ϕprpr 1q pr n
r
n
ϕpr q
1q n!
n
ϕprqpϕprq 1q pϕprq n
implies ϕ
1qq
p q, so that ϕ is numerical.
ϕ r
n
§8. Free Numerical Rings
Recall from Example 4 that a numerical polynomial (over Z) in the variables
x1 , . . . , xk is a formal (finite) linear combination
f pxq ¸
cn1 ,...,nk
x1
n1
xk
,
nk
cn1 ,...,nk
P Z.
Also, a numerical map is a rational polynomial leaving Z invariant. These two
concepts coincide.
Let X be a set, and let EpX q be the term algebra3 based on X. It consists of
all finite words that can be formed from the alphabet
XY
where the symbols
(constants).
"
, , , 0, 1,
nPN
n
and are binary, and
*
,
are unary, and 0 and 1 nullary
n
Definition 3.
The free numerical ring on X is what results after the axioms of a commutative ring with unity, as well as the numerical axioms, have
been imposed upon the term algebra.
Of course, it need be proved that the “free” numerical ring is indeed free
in the usual sense.
Theorem 7.
There is an isomorphism
NRing Z
X
,R
SetpX, Rq,
which is functorial in the numerical ring R.
Moreover,
X
Z
tf P QrX s | f pZX q „ Zu.
3 The
term term algebra is taken from universal algebra; confer Definition II.10.4 of [4].
36
Chapter 1. Numerical Rings
Proof. The numerical axioms, together with the formula for iterated binomial
X
coeHcients, will reduce any element of Z to a numerical polynomial. The
very existence of the numerical ring of numerical polynomials is enough to
guarantee the uniqueness of such a representation. We have thus established
Z
X
tf P QrX s | f pZ q „ Zu.
X
X
From this isomorphism it is evident that Z is free on X, for any set map
X
ϕ : X Ñ R can be uniquely extended to Z by setting
ϕ
¸
cn1 ,...,nk
x1
n1
xk
nk
¸
cn1 ,...,nk
ϕpx1 q
n1
ϕpnxk q .
k
§9. Numerical Transfer
Theorem 8: The Numerical Transfer Principle.
A numerical polynomial
identity ppx1 , . . . , xk q 0 universally valid in Z is valid in every numerical ring.
Proof. From the previous section we have a canonical embedding
Ñ ZZ
ppx1 , . . . , xk q ÞÑ pppn1 , . . . , nk qqpn ,...,n qPZ .
Z
x1 , . . . , xk
k
1
k
k
x
View p as an element of Z x1 ,...,
k . It is the zero numerical map, and therefore
also the zero numerical polynomial.
Example 7.
Recall that a pre-λ-ring (formerly called just λ-ring) is a commutative ring with unity equipped with unary operations λn , n P N, satisfying
the following axioms:
1. λ0 paq 1.
2. λ1 paq a.
3. λn pa
bq ¸
p q n
λp paqλq pbq.
In a numerical ring, the operations λn paq na will evidently satisfy these
axioms.
The definition of a λ-ring (a. k. a. special λ-ring) involves three more axioms,
which are rather cumbersome to state. The reader will believe us when we
claim they are of a polynomial nature, so their verification in a numerical ring
will simply consist in verifying a number of numerical polynomial identities.
As these are valid in Z (for Z itself is well known to be a λ-ring), they will
hold in every numerical ring by Numerical Transfer.
4
§10. The Nilradical
37
§10. The Nilradical
Yet another pleasant property of numerical rings is the following.
In a numerical ring, the congruence
Theorem 9: Fermat’s Little Theorem.
ap a 0 mod p
holds for any prime number p.
Proof. Since
xp x
p
f pxq is a numerical map, it may be written as a numerical polynomial f pxq
But then evidently
ap a pf paq P pR.
P Z x .
Example 8.
The polynomial f can in fact be given explicitly. For when
a P N, we may calculate the number of functions rps Ñ ras as
ap
p
¸
" * k!
k 1
p
k
a
,
k
(
(
where kp denotes a Stirling number of the second kind. Since k! kp counts
the number of onto functions rps Ñ rks, these numbers are all divisible by p,
except in the case k 1, and so
ap a
p
" * p
¸
k! p a
p k
k 2
k
.
It follows from the Numerical Transfer Principle that this formula is valid in
every numerical ring.
4
Theorem 10.
space over Q.
The nilradical of a numerical ring is divisible, and hence a vector
Proof. Let p be a prime and suppose a lies in the nilradical of R. From Fermat’s
Little Theorem p | apap1 1q, from which it inductively follows that
p | apa2
m
pp1q 1q
for all m P N. A large enough m will kill a, and we conclude that p | a.
38
Chapter 1. Numerical Rings
§11. Numerical Ideals and Factor Rings
Let us now make a short survey of numerical ideals and factor rings.
Let I be an ideal of the numerical ring R. The equation
Theorem 11.
r
I
n
r
n
I
will yield a numerical structure on R{I iff
e
n
PI
for every e P I and n ¡ 0.
Proof. The condition is clearly necessary. To show suHciency, note that, when
r P R, e P I, and the condition is satisfied, then
r
e
n
¸
p q n
r
p
e
q
r
n
e
0
r
n
mod I.
The numerical axioms in R{I follow immediately from those in R.
Definition 4.
An ideal of a numerical ring satisfying the condition of the
previous theorem will be called a numerical ideal.
Z does not possess any non-trivial numerical ideals, because
Example 9.
all its non-trivial factor rings have torsion. Neither do the rings Zrm1 s.
4
Theorem 12.
Let R be a (commutative, unital) ring, and let I be an ideal. Suppose
I is a vector space over Q, and that R{I is numerical. Then R itself is numerical, and I is
a numerical ideal.
Proof. Since I and R{I are both torsion-free, so is R, and there is a commutative
diagram with exact rows:
0
/I
/R
/ R{I
/0
0
/ Q bZ I
/ Q bZ R
/ Q bZ R{I
/0
It will suHce to show that R is closed under the formation of binomial coeHcients in Q bZ R. Let r P R. Calculating in in the ring Q bZ R{I yields
r pr 1q pr n
n!
1q
I
r
I
n
.
§12. Finitely Generated Numerical Rings
Since
r I
n
39
in fact lies in R{I, it must be that
r pr 1q pr n
n!
1q
P R,
and we are finished.
That I is numerical follows from the fact that it is a Q-vector space.
The quotient map R
morphism.
Ñ R{I will automatically be a numerical ring homo-
§12. Finitely Generated Numerical Rings
Lemma 4.
If a ring R is torsion-free and finitely generated as an abelian group, its
fraction ring is Q bZ R.
Proof. By the Structure Theorem for Finitely Generated Abelian Groups, R is
isomorphic to some Zn as an abelian group. Let a P Zn . Multiplication by
a is a linear transformation on Zn , and so may be represented by an integer
matrix A. The condition that a not be a zero divisor corresponds to A being
non-singular. It will then have an inverse A1 with rational entries. The inverse
of a is given by
a1 A1 1 P Qn Q bZ R,
where 1 denotes the multiplicative identity of R, considered as a column vector.
Let A denote the algebraic integers in the field K
Lemma 5.
generated over Q, A is finitely generated over Z.
… Q.
If K is finitely
The following theorem, together with its proof, is due to Professor Ekedahl. It classifies completely those numerical rings which are finitely generated
as rings (forgetting the numerical structure).
Before we enter the very technical proof, let us recall from Example 2 that
Zrm1 s inherits a numerical structure from Q. Recall also that products of
numerical rings are numerical, with componentwise evaluation of binomial
coeHcients.
Theorem 13: The Structure Theorem for Finitely Generated Numerical
Let R be a numerical ring which is finitely generated as a ring. There exist
Rings.
unique positive, square-free4 integers m1 , . . . , mk such that
1
1
R Zrm
1 s Zrmk s.
4 A square-free, or simply composite, number is a positive integer that is a (possibly empty) product
of distinct primes.
40
Chapter 1. Numerical Rings
Proof. Case A: R is finitely generated as an abelian group. We first impose the
stronger hypothesis that R be finitely generated as an abelian group.
If r n 0, then, because of Fermat’s Little Theorem, r is divisible by p for
all primes p ¡ n. But in Zn this can only be if r 0; hence R is reduced.
By the lemma above, the fraction ring of R is Q bZ R. As this is reduced and
artinian, being finite-dimensional over Q, it splits up into a product of fields
of characteristic 0.
Case A1: The fraction ring of R is a field. Let us first consider the special case
when the fraction ring Q bZ R is a field, whose ring of algebraic integers we
denote by A. We examine the subgroup A X R of A. Since A „ Q bZ R, an
arbitrary element of A will have an integer multiple lying in R. This means
A{pA X Rq is a torsion group. Also, the fraction ring Q bZ R is finitely generated
over Q, so from the lemma above, we deduce that A is finitely generated over
Z. Because the factor group A{pA X Rq is both finitely generated and torsion,
it is killed by a single integer N, so that
N pA{pA X Rqq 0,
and as a consequence
pA X RqrN 1 s ArN 1 s.
Now let z P A and let p be a prime. The element
z P ArN 1 s pA X RqrN 1 s
can be written z
Theorem(s),
a
,
Nk
P A X R and k P N.
where a
Using Fermat’s Little
pN k qp N k pn
ap a pb
for some n P Z and b P R. Observe that pb belongs to A X R, hence to ArN 1 s,
so that b P A, as long as p does not divide N. We then have
zp z ap
N kp
pa
and hence
Nak Nak
pb
pn
pbqN k apN k
pN k pnqN k
Nak
pnq
N b na
p pNNk b pnna
qN k p N pp 1qk ,
pu zp z P A
k
k
for some u P ArN 1 s, assuming p - N. But then in fact u P A.
Consequently, for all z P A and all suHciently large primes p, the relation
zp z P pA holds, so that zp z in A{pA. Being reduced and artinian, A{pA
may be written as a product of fields, and because of the equation zp z,
these fields must all equal Z{p, which means all suHciently large primes split
§13. Modules
41
completely in A. It is then a consequence of Chebotarev’s Density Theorem5
that Q bZ R Q. Since we are working under the assumption that R is finitely
generated as an abelian group, we infer that R Z.
Case A2:±The fraction ring of R is a product of fields. If the fraction ring of R is
a product Kj of fields,±
the projections Rj of R on the factors Kj will each be
numerical. Hence R „
Rj , with each Rj being isomorphic to Z, according
to the above argument. But Z possesses no non-trivial numerical ideals, so by
Delsarte’s Lemma, R must equal the whole product
R
¹
Rj
¹
Z.
Case B: R is not finitely generated as an abelian group. Finally, we drop the assumption that R be finitely generated as a group, and assume it finitely generated as a ring only. Because of the relation p | r p r, R{pR will be a finitely
generated torsion group for each prime p. It will then have Krull dimension
0, and it follows from Chevalley’s Dimension Argument that dim Q bZ R 0,
so that Q bZ R is a finite-dimensional vector space over Q. Only finitely many
denominators are employed in a basis, so there exists an integer M for which
RrM 1 s is finitely generated over ZrM 1 s.
We can now more or less repeat the previous argument. RrM 1 s will still
be reduced, and as before, Q bZ RrM 1 s will be finite-dimensional, hence a
product of fields, and we may reduce to the case when Q bZ RrM 1 s is a
field. Letting A denote the algebraic integers in Q bZ RrM 1 s, the factor group
A{RrM 1 s will be finitely generated and torsion, and hence killed by some integer, so that again we are lead to RrN 1 s ArN 1 s. As before, we may draw
the conclusion that Q bZ R Q, and consequently that R ZrN 1 s. This
concludes the proof.
§13. Modules
A most elegant application of the Structure Theorem is the classification of
torsion-free modules.
Lemma 6.
Consider a ring homomorphism ϕ : R
torsion-free, then Ker ϕ will be a numerical ideal.
Ñ S. If R is numerical and S is
Proof.
n!ϕ
if r
r
n
ϕ
P Ker ϕ and n ¡ 0. Thus
r
n
n!
r
n
ϕprpr 1q pr n
1qq 0,
P Ker ϕ, which is then numerical.
5 (A special case of) Chebotarev’s Density Theorem states the following: The density of the
primes that split completely in a number field K equals |GalpK1 {Qq| . In our case, this set has density
1.
42
Chapter 1. Numerical Rings
Let M be a torsion-free module over the numerical ring R, with module
structure given by the group homomorphism
µ : R Ñ End M.
We have the following commutative diagram:
0
0
w
ww
w
w
ww
w{ w
/R
/ R{ Ker µ
/0
s
s
s
µ
ss
ss
s
sy
End M
/ Ker µ
The group End M is torsion-free, so, by the lemma, Ker µ is a numerical ideal.
Therefore M will in fact be a module over the numerical ring R{ Ker µ.
Let us now also assume that End M is finitely generated (as a module) over
Zrn1 s for some integer n. Because Zrn1 s is a noetherian ring, End M is a
noetherian module. The submodule R{ Ker µ is finitely generated as a module
over Zrn1 s, and therefore also as a ring, so by the Structure Theorem,
1
1
R{ Ker µ Zrm
1 s Zrmk s,
for square-free, positive integers mj . The module M will split up as a direct
sum
M M1 ` ` Mk ,
1
with each Mj a torsion-free module over Zrm
j s. Because these rings are principal, the modules Mj are in fact free, and we have proved:
Theorem 14.
Consider a module M over a numerical ring. Suppose M is torsionfree and finitely generated over Zrn1 s for some integer n. There exist positive integers
mj , rj such that
1 rk
1 r1
M Zrm
1 s ` ` Zrmk s
as a module over
1
1
Zrm
1 s Zrmk s.
§14. Exponentiation
When A is a ring, the symbol
?0
A
shall denote its nilradical.
§14. Exponentiation
43
Let R be a numerical ring, and consider a (commutative,
algebra
?A 0, givenunital)
by the (finite)
A over it. There is an induced exponentiation on 1
binomial expansion:
8̧ r p1 xqr xn .
n
n 0
The numerical axioms imply the following properties.
I. p1
II.
III.
IV.
V.
xqr p1
xqs
p1 xqr s .
p1 xqr s p1 xqrs .
p1 xqr p1 yqr p1 xqp1 yq r .
p1 xq1 1 x.
p1 xqr 1 rx modp?0q2 .
?
Exponentiation will thus make the abelian group p1 A 0, q into an R-module.
Indeed, property III shows that exponentiation by r gives an endomorphism
εpr q of the group, and properties I, II, and IV show that
?0, q
εA : R Ñ Endp1
A
is a unital ring homomorphism.
The module structure is natural in the following sense. Given two algebras
A and B and an algebra homomorphism ϕ : A Ñ B, the subsequent diagram
commutes for any r P R:
?0
A
1
ϕ
1
?
B
pq/
εA r
1
?0
A
ϕ
0
/
pq 1
εB r
?
B
0
Let us now reverse the procedure.
Theorem 15: The Binomial Theorem. Let R be a commutative, unital ring.
• If R is numerical, the equation
p1
defines a module structure on p1
in addition satisfies
p1 xqr
xq
r
8̧
n 0
r n
x
n
(4)
?0, q, which is natural in R-algebras A, and
A
1
?
rx modp 0q2 .
(5)
44
Chapter 1. Numerical Rings
?
• Conversely, if there is a natural module structure on p1 A 0, q for all R-algebras
A, satisfying (5), then there is a (necessarily unique) numerical ring structure on R,
fulfilling the equation (4).
Proof. There remains to establish the second part. Let a natural module structure be given, and consider
?0, q,
εA : R Ñ Endp1
where A Rrts{ptn
A
q, and n is some (large) number. We have
εpr qp1 tq p1 tqr a0 a1 t an tn ,
1
and clearly the coeHcients
ak are independent of n. Therefore, we may without
ambiguity define kr ak . This will make the identity (4) hold in A, and then
it will hold everywhere by naturality.
The axioms for a numerical ring should now be immediate, as they are
simply direct translations of the module axioms. For example, identification
of the coeHcients of tn in
8̧ r
8̧ s
i
j
i 0
i
t
j 0
j
t
p1 tq p1 tq p1 tq r
s
r s
8̧
n 0
r
s
n
tn
proves axiom I. (Proving III will of course involve the polynomial ring in two
variables.)
And this little “treatise upon the Binomial Theorem” closes the chapter
on numerical rings.
Chapter 2
MULTI-SETS
Är Du en Enhet eller delar?
Jag bäfvar, mod och sansning felar,
Min fråga giör mig stel och stum.
Hedwig Charlotta Nordenflycht,
Ode i Anledning af Exod. XXXIII: Cap. v. 18. 20.
och XXXIV: Cap. v. 5. 6.
The text will be pervaded by the use of multi-sets, and we develop here their
theory from scratch. This we do partly to fix notation, and partly because
some concepts we need are possibly not standard. We certainly lay no claims
of originality upon the theory explored in this chapter. It is conceivable, and
even very likely, that all the results of this chapter can be found somewhere in
the literature.
After giving the basic definitions, we propose to answer the following question: What would be the natural arrows of a category of multi-sets? The
non-existence of a definite answer does not depend on a lack of suggestions.
Dr. Salomonsson’s thesis [20] presents a plethora: multijections, multi-maps,
maps, and bimultijections. He finally decides to build the multi-set category
with multijections as the basic arrows, a choice we believe is less suited to our
purposes. We have settled on the latter kind, bimultijections, as giving the most
natural theory, and in the process renamed them multations. Our conviction
that this is the correct choice stems from the tight connection that is seen to
exist between multations and divided powers.
One advantage of using multijections is that they allow for a unique composition. We have found it expedient to drop that requirement. The very
nature of multi-sets, with their repeated elements, seems to exclude unambiguous composition in the usual sense. The solution we have accepted (which
is so natural, it might almost be termed canonical) is to “sum over all possible
compositions”.
§1. Multi-Sets
Definition 1.
A multi-set is a pair
M
p#M, degM q,
45
46
Chapter 2. Multi-Sets
where #M is a set and
ÑZ
degM : #M
is a function, called the degree, or multiplicity. The underlying set #M is
called the support of M.
We call
degM a
the degree or multiplicity of an object a P #M; it counts the “number of times
a occurs in M”. The degree of the whole multi-set M we define to be
¹
deg M
P
pdeg xq!.
x #M
It might conceivably be convenient to have the degree function defined on
the whole set-theoretic universe. A multi-set may then equivalently be defined
as a function
M degM : Set Ñ N
vanishing outside some set. The support is given by
#M
tx | degM pxq 0u.
In order to give a multi-set, it suHces to specify such a degree function.
Definition 2.
The cardinality of M is
|M | ¸
P
1
x M
¸
P
deg x.
x #M
The cardinality counts the number of elements with multiplicity. We tacitly assume all multi-sets under discussion to be finite, as these are the only
ones we will ever need.
Example 1.
The multi-set ta, a, bu has cardinality 3 and support ta, bu. We
have deg a 2, deg b 1, and deg c 0.
4
It is now an easy matter to generalise the elementary set operations to
multi-sets.
Definition 3.
The union A Y B of A and B is
degAYB
Definition 4.
The disjoint
maxpdegA , degB q.
union1
degA\B
1 Please
A and B.
A \ B of A and B is
degA
degB .
note that [20] employs a diverent definition of disjoint union, and calls this the sum of
§1. Multi-Sets
Definition 5.
47
The intersection A X B of A and B is
degAXB
Definition 6.
The relative complement AzB of B in A is
degAzB
Definition 7.
degA degB : #A #B Ñ Z
.
A is a sub-multi-set2 of B, written A „ B, if
degA
¤ degB
(element-wise inequality).
Definition 9.
maxpdegA degB , 0q.
The direct product A B of A and B is
degAB
Definition 8.
minpdegA , degB q.
The power multi-set of A is
2A
tB | B „ Au,
where every sub-multi-set of A is counted “according to multiplicity”.
In other words, the classical formula |2A | 2|A| will still be valid.
Given A tx, x, yu and B tx, y, zu, we have
Example 2.
A Y B tx, x, y, zu
A \ B tx, x, x, y, y, zu
A X B tx, yu
AzB txu
BzA tzu
A B tpx, xq, px, xq, py, xq, px, yq, px, yq, py, yq, px, zq, px, zq, py, zqu
2A
t∅, txu, txu, tyu, tx, xu, tx, yu, tx, yu, tx, x, yuu.
4
Recall that the Principle of Inclusion and Exclusion, in one form, states
the following: If f and g are functions such that
¸
„
X Y
2 Some
people would say multi-subset.
f pX q gpY q,
48
Chapter 2. Multi-Sets
then
¸
f pY q „
p1q|Y ||X | gpX q.
X Y
Of course, X and Y are here limited to be sets, but a generalisation to multisets is immediate.
Theorem 1: The Principle of Inclusion and Exclusion.
Let S be a set with
n elements. Consider functions f defined on multi-sets with support included in S and
cardinality n; and functions g defined on the power set of S. If
¸
f pAq gpX q,
„
|A|n
#A X
then
¸
|A|n
¸
f pAq #A Y
„
p1q|Y ||X | gpX q.
X Y
Proof. Calculate
¸
„
p1q|Y ||X | gpX q X Y
¸
„
p1q|Y ||X |
X Y
¸
„
|A|n
¸
„
|A|n
f pAq
#A X
p1q|Y | f pAq
#A Y
¸
„ „
p1q|X | ,
#A X Y
and note that the inner sum vanishes, unless #A Y .
§2. Multations
Let A and B be multi-sets of equal cardinality. A multation µ : A Ñ B is a
pairing of their elements. We shall write multations as two-row matrices,
with the elements of A on top of those of B, the way ordinary permutations
are usually written:
a1 an
µ
b1 bn
The order of the columns is of course irrelevant.
Observe that µ is not a “function” from A to B, since identical copies of
some element of A may very well be paired ov with distinct elements of B. It
will, however, be a sub-multi-set of AB, with the property that every element of
A occurs exactly once as the first component of a pair in µ, and each element
of B exactly once as a second component. The degree degµ pa, bq counts the
number of times a P A is paired ov with b P B.
§3. Confluent Products
Given a multation
49
a1
b1
a1
b1
a2
b2
...
...
a2
b2
...
,
...
aj
, we shall adopt the perspective of
bj
with mj appearances of the column
viewing it as a formal product
rm1 s rm2 s
a1
b1
a2
b2
...
of divided powers3 .
There exist two multations from the multi-set ta, a, bu to itself,
Example 3.
namely:
a
a
a
a
b
b
r2s a
a
b
b
a
a
a
b
b
a
a
a
a
b
b
.
a
The degree of pa, bq is 0 with respect to the first of these, and 1 with respect to
the second.
4
§3. Confluent Products
Some heavy notation will inevitably come into play when writing a thesis.
We here describe some shorthand, which will be used extensively for the remainder of the text.
Let A be a multi-set, and let M be a module. Consider elements xa P M
indexed over the support of A. Define the confluent product over A as
ä
P
xa
a A
¹
P
rdeg as P ΓpM q.
xa
a #A
For example, we have
x d x d y xr2s y.
We may further abbreviate
xr A s
and similarly
xA
ä
P
xa
a A
¹
P
a A
¹
P
rdegA as ,
xa
a #A
xa
¹
P
degA a
xa
.
a #A
3 A divided power xrns should be thought of as x . We shall later discuss divided power algebras,
n!
but some previous familiarity with these structures will be assumed of the reader.
n
50
Chapter 2. Multi-Sets
This latter product is defined in any algebra.
A special case of this practice arises when µ : A Ñ B is a multation, and the
variables xba have been doubly indexed over the supports of B and A simultaneously. We then have
xr µ s
and
xµ
ä
pa,bqPµ
¹
pa,bqPµ
xba
rdegµ pa,bqs
¹
xba
P
P
,
a #A
b #B
xba
¹
p q
degµ a,b
xba
P
P
.
a #A
b #B
As a special case of that practice, we have, for quantities xa and yb indexed
by #A and #B, respectively,
pxyqrµs ä
pa,bqPµ
xa y b
¹
P
P
pxa yb qrdeg pa,bqs .
µ
a #A
b #B
Finally, consider a multation µ : A Ñ B and variables xb
support of B. We then let
xbrµs
â
P
ä
xb
p qPµ
a # A a, b
So for example, if
µ
then
xbrµs
1
1
P
1
1
â
P
a #A
P M indexed by the
ΓdegA a pM q ΓA pM q.
1 2 2
,
2 2 3
xr12s x2 b x2 x3 .
§4. The Multi-Set Category
Let µ : B Ñ C and ν : A Ñ B be two multations, where |A| |B| |C | n.
Their composition µ ν is found by identifying the coeHcient of xµ yν in the
equation
rn s
¸
b xcb
c P
P
b #B
c #C
rns
¸
a yba
b P
P
a #A
b #B
rns
¸
aP#A
P
P
xcb yba
a c .
b #B
c #C
In somewhat fancier language, we have defined a formal multiplication of
columns:
$ 1 '
& a
if b b1 ,
b
a
c
c
b
'
%
0
if b b1 .
§4. The Multi-Set Category
51
This makes the free module of columns into an algebra A. Multation composition is then given by the formula
urns vrns
puvqrns ,
an operation which we shall later have occasion to baptise the product multiplication on Γn pAq.
For example, to find the composition
Example 4.
c
a
d
a
d
b
a a
c d
b
,
d
we use the equation
r3s
c
a
xac
d
a
xad
a
a
xac yca
d
b
xbd
xad yda
a
a
xad ydb
b
a
r3s
a
c
yca
a
d
yda
ydb
r3s
a
b
xbd yda
b
d
xbd ydb
b
b
.
Identification of the coeHcients of
xac xad xbd yca yda ydb
yields
c
a
d
a
d
b
a a
c d
b
d
2 aa
a
a
b
b
a
a
a
b
b
.
a
Similarly, by picking the coeHcients of
xac x2ad yca y2da ,
we obtain
c
a
d
a
d
a
a a
c d
a
d
3 aa
a
a
a
.
a
4
There is a simpler, combinatorial rule for calculating the composition.
Namely, the composition of two divided power products is found by “summing over all possibilities of composing them”:
b1
c1
bn
cn
a1
b1
an
bn
¸
σ
a1
cσp1q
an
cσpnq
,
where the sum is to be taken over all permutations σ : rns Ñ rns such that
bj bσpjq for all j. We leave it to the reader to check the accuracy of this rule.
52
Chapter 2. Multi-Sets
Computing according to this device, we have
Example 5.
c d
a a
d
b
a
c
a
d
b
d
c
a
d
a
d
b
a
a
a
a
b
b
2
2
a
c
a
d
b
d
a
a
a
b
b
a
a
a
a
b
b
a
r2s a
a
b
b
a
a
a b
a b
a a
a b
b
,
a
and similarly
c
a
d
a
d
a
a a
c d
a
d
r2s
c
a
d
a
r2s
a
c
a
d
1 c d d
21 ac
2 a a a
1
2 aa aa aa
4
3
r3s
a
a
3
a
a
a
a
a
d
a
d
a
.
a
4
The identity multation (“identitation”) ιA of a multi-set A is the multation
in which every element is paired ov with itself. It is clear that composition is
associative and that the identity multations act as identities.
Definition 10.
The nth multi-set category has as objects the multi-sets of
cardinality exactly n. Given two multi-sets A and B, the arrow set MSetn pA, Bq
is the free module generated by the multations A Ñ B.
§5. Multi-Sets on Multi-Sets
Unfortunately, we shall have to push things one step further, and deal with
multi-sets supported by multi-sets, which may sound like rather a baroque
consideration. But the situation is not nearly as unpleasant as it was originally,
before Professor Franjou graciously helped us tidy things up a bit, for which
we are humbly grateful.
Definition 11.
Let M be a multi-set. A multi-set supported in M is a
multiset supported in the set
M#
px, kq P #M Z
(
1 ¤ k ¤ deg x .
§6. Partitions and Compositions
53
Example 6.
Let M be the set ta, b, cu and N the multi-set tx, x, yu. There are
three multi-sets with support M and cardinality 4:
ta, a, b, cu, ta, b, b, cu, ta, b, c, cu.
Likewise, since
N#
tpx, 1q, px, 2q, py, 1qu,
there are three multi-sets with support N and cardinality 4:
tpx, 1q, px, 1q, px, 2q, py, 1qu, tpx, 1q, px, 2q, px, 2q, py, 1qu, tpx, 1q, px, 2q, py, 1q, py, 1qu.
4
When speaking of multi-sets supported in a multi-set M, we will let “degree over M” stand for “degree over M # ”.
The three multi-sets above supported in N have all degree 2
4
Example 7.
over N.
§6. Partitions and Compositions
Partitions and compositions of multi-sets will come into play when we investigate operads. The generalisation from the case of sets should be straightforward.
Definition 12.
of X if
Let X be a multi-set. The multi-set E constitutes a partition
§
P
Y
X.
Y E
We let
Par X
denote the set of partitions of X.
Example 8.
The four partitions of t1, 1, 2u are
t1, 1, 2u t1, 1u \ t2u t1u \ t1, 2u t1u \ t1u \ t2u.
4
Definition 13.
Let X be a multi-set. The multation ω : A
an A-composition of X if
§
ωpaq X.
Ñ B constitutes
P
a A
We let
ComA X
denote the set of A-compositions of X.
54
Chapter 2. Multi-Sets
The ten ta, a, bu-compositions of t1, 1, 2u are
Example 9.
a
t1, 1, 2u
a
t1, 1u
a
t2u
b
,
∅
a
t1, 2u
a
t1u
b
,
∅
a
t 1u
a
t1u
a b
,
∅ ∅
a
∅
a
∅
b
t1, 1, 2u ,
a
a
t1, 1u ∅
b
t2u ,
a
t1, 2u
b
t1u ,
b
t2u ,
a
∅
a
t2u
(Each row corresponds to a partition above.)
a
t1 u
a
∅
a
t2u
b
t1, 1u ,
a
∅
a
t1u
b
t1, 2u ,
b
t1u .
4
Chapter 3
MAZES
Dedalus genom sin konst och sitt snille vida beryktad
Bygde det opp; han förvirrar de ledande märken och ögat
I villfarelse för ibland skiljaktiga vägar.
Så på de Frygiska fält, man ser den klara Meandros
Leka. I tveksamt lopp han rinner och rinner tillbaka,
Möter sig ofta sjelf och skådar sin kommande bölja,
Och nu till källan vänd, nu åt obegränsade hafvet,
Rådvill öfvar sin våg. Så fyllas af Dedalus äfven
Tusen vägar med irrande svek: Knappt mäktar han sjelf att
Hitta till tröskeln igen. Så bedräglig han boningen danat.
Ovidius, Metamorphoses
Once upon a time1 , we introduced2 the concept of mazes. A maze from the set
ta, b, c, du to the set tw, x, y, zu (say) is something like the following diagram:
a* ? k / w
** ??? G J ** l??p ** ?? r
b / **n / x // ** /// ***
// **
c /q * / y o/ * /m
//** //** / d ts / z
where k, l, m, n, o, p, q, r, s, t are scalars. The reason for applying the name maze
to such a contraption should be apparent from the picture. This is not to
say that other names have not been suggested. “Why not quiver,” we were
once asked, “for something that obviously contains a lot of arrows?” Because
quivers have already been invented.
The labyrinth category is the category with finite sets as objects, and mazes
as arrows. It might at first seem a puzzling object. Why would anyone be
1 This
2 Syn.:
was back in 2009, in our licenciate thesis.
discovered, invented.
55
56
Chapter 3. Mazes
interested in a category with such strange arrows (and an even stranger law of
composing them)? The answer is that the labyrinth category provides a very
natural, one might even say canonical, means of encoding endofunctors of the
category Mod.
When defining the labyrinth category and exploring its properties, we remark that the base ring B need not be commutative. The existence of a unity
is required, as always, but otherwise it may be of quite an arbitrary nature.
§1. Mazes
Let X and Y be finite sets. A passage from x P X to y P Y is a (formal) arrow p
from x to y, labelled with an element of B, denoted by p. This we write as
p : x Ñ y,
or
p
x
/y,
though we shall frequently forget the bar over p when no confusion is likely
to arise.
Definition 1.
A maze from X to Y is a multi-set of passages from X to Y .
It is required that there be at least one passage leading from every element of
X, and at least one passage leading to every element of Y . (We, so to speak,
wish to prevent dead ends from forming.)
Because a maze is a multi-set, there can (and, in general, will) be multiple
passages between any two given elements.
Example 1.
The following is a maze from tzu to tx, yu:
w; ; x
wwwww z b Q
EEEE ,
c " y
a
whereas the following is not:
w; ; x
wwwww z b .
a
y
4
Example 2.
It is perfectly legal to consider the empty maze ∅ Ñ ∅. It is the
only maze into or out of ∅, and is the only maze having no passages.
4
Definition 2.
multi-sets.
We say P : X
ÑY
is a submaze of Q : X
ÑY
if P
„ Q as
§2. The Labyrinth Category
Example 3.
57
The following is a submaze of Q above:
w; x
ww
z
P
EEEE .
c " y
a
4
Definition 3.
Let P : X Ñ Y be a maze, and consider subsets X 1 „ X and
Y 1 „ Y . The restriction of P to X 1 Ñ Y 1 is the maze (if indeed it is one)
containing only those passages of P that begin in X 1 and end in Y 1 . It will be
denoted by

P 1 1 .
X
ÑY

Note that P X 1 ÑY 1 is not a submaze of P (unless X 1
Example 4.
With Q as above, we have
X and Y 1 Y ).
; x
ww ;
z wwwbw .
a

Q
tzuÑtxu
4
Restrictions may not always exist, as in the following example.
Example 5.
If
R
x
y
a
b
/x
/y
,

then the restriction RtxuÑtyu does not exist (it is not a maze). In such a case,
it will be convenient to define the symbol

RtxuÑtyu 0.
4
§2. The Labyrinth Category
Passages p : y Ñ z and q : x
where the other begins.
Ñ y are said to be composable, because one ends
Definition 4.
If P : Y Ñ Z and Q : X Ñ Y are mazes, we define the
cartesian product P b Q to be the multi-set of all pairs of composable passages:
PbQ
! zo
p
y , yo
q
x
zo
p
y
PP ^
yo
q
x
PQ
)
.
58
Chapter 3. Mazes
For a sub-multi-set U
„ P b Q, we shall write
U „PbQ
to indicate that the projections on P and Q are both onto.
Note that such a set U can be naturally interpreted as a maze itself, namely
!
zo
pq
zo
x
p
q
y , yo
x
PU
)
.
Observe carefully the order in which p and q occur. The surjectivity condition
on the projections will prevent dead ends from forming.
When writing P b Q, we will sometimes refer to the cartesian product, and
sometimes the maze which thus is naturally associated therewith, and hope
the circumstances will make clear which is meant.
Example 6.
Consider the two mazes
x eJJa
P {vv z ,
y b
x
ytt
Q z cHH .
d y
c
Their cartesian product is
PbQ
" xo
a
xo
a
z , zo
z , zo
d
c
x
,
y
,
yo
b
yo
b
z , zo
z , zo
d
y
c
,
x
*
,
which we identify with the maze
x oaD
ac
x
z
zzDDD
D
}zbc
yo
y
adDDzzz
.
bd
4
We now define the composition of two mazes. As for multi-sets, this composition will not in general be a maze, but rather a sum of mazes, and living
in the free module generated by those.
Definition 5.
formal sum
The composition or product of the mazes P and Q is the
PQ ¸
„b
U P Q
U.
§2. The Labyrinth Category
Let P and Q be as above. Their composition is
Example 7.
PQ ac /
xC
=x
CC
CC{bc
{C
{{
{{ adC/!
y
y
bd
ac
x
y
59
{bc
{{
{
{{
bd
/= x
/y
ac
x
y
bd
/x
/y
ac /
xC
x
CC
C
CC
adC!
/y
y
xC
=x
CC
C
bc
C{
{{C
{
ad
C
{
{
!
y
y
ac /
xC
=x
CC
CC{bc {C
{{ adC {
{
!
y
y
bd
xC
=x
CC
CC{bc
{C
{{
{{ adC!/
y
y
.
bd
4
That composition is associative follows from the observation that pPQqR
and P pQRq both equal
¸
W.
„b b
W P Q R
There exist identity mazes
IX
¤!
P
x
1
/x
)
.
x X
Definition 6.
The labyrinth category Laby has as objects the finite sets.
Given two sets X and Y , the arrow set LabypX, Y q is the module generated by
the mazes X Ñ Y , with the following relations imposed.
I.
PY
!
/
0
)
0,
for any multi-set P of passages.
II.
PY
PY
"
!
a b
a
/
/
*
)
PY
"
b
/
*
PY
#
a
b
/
/
+
,
for any multi-set P of passages.
The second axiom may be generalised by means of mathematical induction
to yield the following elementary formulæ.
60
Chapter 3. Mazes
Theorem 1.
PY
PY
In the labyrinth category, the following two equations hold:
#
n
¤
°n
/
a
i 1 i
"
ai
/
+
*
¸
€ „rns
¤
PY
P
i 1
I
„rns
i I
∅ I
¸
"
ai
#
p1qn|I | P Y *
/
°
P /
a
i I i
+
.
§3. Operations on Mazes
Passages p : x Ñ y and q : x Ñ y sharing starting and ending points, will be said
to be parallel.
Definition 7.
allel passages.
The maze P is called simple, if it contains no (pair of) par-
By means of the labyrinth axioms I and II, any maze can be written as the
sum of simple mazes.
Definition 8.
The mazes P, Q : X Ñ Y are similar, if, for any x
y P Y , there are as many passages x Ñ y in P as there are in Q.
Example 8.
P X and
The following two mazes are similar:
w; ; x
wwwww z E b ,
EEE c " y
w; ; x
wwwww z E e .
EEE "y
f
a
d
4
Definition 9.
product is
Let P : Y
PeQ
Ñ Z and Q : X Ñ Y
"
°
zo
pq
x
where the sum is taken over all pairs z Ð y
able passages.
p
be mazes. Their functional
z P Z, x P X
*
Let
x eJJa
P {vv z ,
y b
P P and y Ðq x P Q of compos
The functional product is always a simple maze.
Example 9.
,
x
ytt
Q z cHH .
d y
c
§3. Operations on Mazes
61
Then the functional product P e Q coincides with the cartesian product P b Q,
computed earlier. The functional product
QeP
ca
zo
db
z
,
however, divers from the cartesian product
QbP
o
zo
ca
z
.
db
4
The following formula relates the cartesian and functional products.
¸
Theorem 2.
„b
V
V P Q
¸
„e
W.
W P Q
Proof. Consider a submaze W „ P e Q. It has at most one passage running
between any two given elements. We then consider the set
EpW q tV
„ P b Q | D rx Ñ z s P V Ø D rx Ñ z s P W u .
Apply the first formula of Theorem 1 to each passage of EpW q to show
¸
V W.
Pp q
V E W
This proves the theorem.
Definition 10.
Let P be a maze, and let a be a scalar. The functional scalar
multiplication of P by a is the maze a d P obtained from P by multiplying the
labels of all passages by a:
adP
!
yo
ap
x
yo
p
x
PP
)
.
Naturally, one should distinguish between left and right multiplication,
but we shall only employ this operation in the case of a commutative base
ring.
Definition 11.
al sum is
P`Q
Let P, Q : X
!
x
p q
/y
Ñ Y be similar, simple mazes. Their function-
x
p
/y
P P,
x
q
/y
PQ
)
.
62
Example 10.
Chapter 3. Mazes
Plainly,
a c w; x
c w; x
w; x
ww ww ww
zE
z
z
EEE ` EEEE EEEE .
"
"
"
b
d
b d y
y
y
a
4
Applying some induction to Theorem 1 yields the following formula for
the functional sum.
Theorem 3.
Let P1 , . . . , Pn be similar, simple mazes, and let the passages of Pi be
pi1 , . . . , pim (where it is understood that the passages p1j , . . . , pnj all run in parallel). Then
P1 ` ` Pn
¸
tpij | pi, jq P K u,
K
where the sum is taken over all K
ponent is onto.
„ rns rms such that projection on the second com-
These functional operations will later be shown to have very natural interpretations, which will account for the epithet “functional”.
Definition 12.
maze
Let P : X
Ñ Y be a maze in LabypBq. The dual of P is the
P : Y
ÑX
in LabypB q, obtained by reversing all passages in P.
Of course, the dual will also belong to LabypBq, but it is more natural to
let it lie in LabypB q. Letting denote the multiplication of B (the opposite
multiplication of B), and also the corresponding induced multiplication on
the labyrinth category LabypB q, we have the equation
pPQq Q P .
The dual thus provides an isomorphism of categories
LabypBq
LabypB q.
§4. The Quotient Labyrinth Categories
Everything up to this point makes sense for an arbitrary base ring. In order
to construct quotient categories of Laby, we need the assumption that B be
numerical.
Let P : X Ñ Y be a maze. We shall consider multi-sets A supported in P
(recall this notion from Chapter 2). This is intended to capture the informal
notion that A : X Ñ Y is the maze that results after certain passages of P have
been multiplied.
§4. The Quotient Labyrinth Categories
63
When A is a maze, we let EA denote the maze
EA
!
¤
rp : xÑysPA
1
x
/y
)
,
in which all passages of A have been reassigned the label 1.
While the full labyrinth category encodes arbitrary module functors, the
following quotient category will be seen to encode numerical functors (of degree
n).
Definition 13.
The category Labyn is the quotient category obtained from
Laby when the following relations are imposed:
III.
P
0,
whenever P contains more than n passages.
IV.
P
¸ ¹
P
|A|¤n
#A P p P
p
EA ,
degA p
for all mazes P.
An instance of the fourth axiom is the following:
Example 11.
qq8
a b q
MMM
1
1
&
a
b
1
q8 b qqqqq8 MMM1 &
1
1
a
2
q8 MqMqM& 1
1
a
1
8
b q1qq M
1
MMMMM& .
2
1 & 4
It may be proved that, over the integers, Axiom IV is redundant, as it
follows from Axiom III.
Definition 14.
pure maze.
A maze of which all passages carry the label 1 is called a
The pure mazes evidently generate the category Labyn . Not as evident,
but equally true, is that they actually consitute a basis. This will be proved
presently.
We shall have reason to impose upon the labyrinth category yet another
axiom. This will be for encoding quasi-homogeneous functors (of degree n).
64
Chapter 3. Mazes
The category Labyn is the quotient category
Definition 15.
Labyn {pL X Labyn q,
where L is the ideal of Q bZ Labyn generated by the following elements:
V.
an P a d P,
for any pure maze P and a P Q bZ B.
Using the fourth axiom, this fifth axiom can be equivalently written
n
a P
¸
a
A.
A
|A|¤n
#A P
Implicit in the fifth axiom is the existence of an inclusion of categories
Labyn
„ Q bZ Labyn ,
which is not quite justified at this point, unless we take it on faith (which we
do) that the category Labyn is free, and hence torsion-free.
Example 12.
The fifth axiom considers the ideal generated in Q bZ Labyn ,
rather than just Labyn . The slightly sharper requirement will first make a
diverence in degree 4. Consider, for example, Z Laby4 . Dividing out by the
ideal generated in Laby4 by elements of the form
a4 P a d P
makes it possible to prove
0
:
I
v
vv1v 2 HH 1 H 1 H$
1
v
1 vv: :
vvvv v 1 2
HHHHH1H H $
1 H$ 1 vv:
vv
2
HHHHH1H 12 HH $
1 $ 1 vv:
v
vHHH1
H$
,
whereas we shall be needing a stronger statement. By allowing the full force
of Axiom V, we make it possible to divide by 2, and thus establish
:0 I
v
v
vv1
H 1 HH 1 H$
1
1 vv: :
vvvvv v 1 H
HHHH1H H $
1 H$ 1 vv:
v
HvHHHH1H
1
HH $
$ 1 vv:
v
6 vHHH1
H$
.
4
Chapter 4
MULTI-SETS
VERSUS
MAZES
I leave to the various futures (not to all) my garden of forking paths.
Jorge Luis Borges, The Garden of Forking Paths
We propose to investigate how the multi-set and labyrinth categories are related. A functor in one direction is readily found; viz. the Ariadne functor
An : Laby Ñ MSetn ,
so called because it leads the way out of the labyrinth. In the case of a numerical base ring, it turns out to factorise, yielding a functor
An : Laby Ñ Labyn
Ñ Labyn Ñ MSetn .
We will later see that multi-sets encode strict polynomial functors, while
mazes encode numerical ones. The Ariadne functor will then correspond to
the forgetful functor from the former functor category to the latter.
Does there exist a functor in the other direction? No, at least not in general. It is, however, possible to slightly tweak the labyrinth category in some
rather non-invasive way, which will enable defining a functor in the reverse
direction. This is the Theseus functor Tn , going into the labyrinth.
For the purposes of this chapter, we shall modify slightly the definitions
of the categories of multi-sets and mazes by adjoining to them direct sums, thus
making them additive. This can be done in a way that is not only well-known,
but universal (left adjoint to a forgetful functor, and so on); hence we omit the
details. It will make no diverence for linear functors out of these categories,
as we shall later have occasion to consider.
One of the reasons for keeping the original definition is that, sometimes,
it will be nice to know that the categories Labyn and MSetn possess finite
skeleta. Not actually helpful, perhaps, but still nice to know.
Finally, we pay tribute to Dr. Dreckman and Professors Baues, Franjou, and
Pirashvili, as their article [1] has provided a wonderful source of inspiration for
our work. Their main object of study Sur, the category of sets and surjections,
may rightly be called the early ancestor of the labyrinth category.
65
66
Chapter 4. Multi-Sets versus Mazes
§1. The Ariadne Functor
For the duration of this section, let n be a fixed natural number.
Let P be a maze. Consider the following sum of multations:
¸
An pP q ä
rp : xÑysPA
|A|n
#A P
x
p
y
¸
1
degA P
# A P
|A|n
¹
rp : xÑysPA
x p
.
y
This will provide a functor from Laby to MSetn , which we now set out to
prove. It is clear that An pP q 0 if a single passage of P is labelled 0. Now to
show that
An P Y
An P Y
"
u
!
/v
a b
a
u
/v
)
*
An P Y
Denote
Since
µ
"
An P Y
u
An P Y
An P Y
An P Y
!
u
"
u
/v
a b
a
b
#
u
a
b
/v
/v
/v
/
*
)
*
+
"
u
b
ä
rp : xÑysPA
¸
ņ
| | m 1 #A P
A n m
¸
ņ
| | m 1 #A P
A n m
¸
ņ
| | ņ
¸
/v
An P Y
#
a
u
b
/v
/
+
.
x
.
y
p
m 1 #A P
A n m
*
µ d pa
µda
m
µdb
m
¸
¥ | | m 1 i j m #A P
i, j 1 A n m
bq
m
rms u
v
rms u
v
rms u
v
µ d ai b j
ris rjs u
v
u
v
,
this relation follows from the equation
px
yqrms
xr m s
y rm s
¸
¥
xris yrjs .
i j m
i, j 1
Hence An gives a well-defined map on the mazes of the labyrinth category.
We now prove that it is in fact a functor.
§1. The Ariadne Functor
67
The formulæ
Theorem 1.
An pX q An pP q ¸
à
|A|n
A
#A X
ä
x
y
p
rp : xÑysPA
|A|n
#A P
provide a linear functor
Proof. Let P : Y
An : Laby Ñ MSetn .
Ñ Z and Q : X Ñ Y be two mazes. We wish to show that
An pP q An pQq ¸
ä
rp : yÑzsPA
|A|n
#A P
¸
¸
¸
„b
| |
R P Q #C R
C n
¸
An „b
¸
ä
rq : xÑysPB
|B|n
#B Q
q
x
y
ä
rr : xÑzsPC
„b
|C |n
#C P Q
y
p
z
r
x
z
ä
rr : xÑzsPC
x
z
r
R An pPQq.
R P Q
The dubious step here is the third one, which follows from the equation
¸
rp : yÑzsPP
rn s
y p
z
¸
rq : xÑysPQ
r ns
x q
y
¸
rp : yÑzsPP
rq : xÑysPQ
rns
x ,
pq
z after noting that restricting attention to monomials pA qB with #A P and
#B Q in the left-hand side corresponds to considering ppqqC with C „ P b Q
in the right-hand side.
Definition 1.
This functor is called the nth Ariadne functor.
Theorem 2.
The Ariadne functor factors through the quotient category Labyn , producing a functor
An : Labyn Ñ MSetn .
68
Chapter 4. Multi-Sets versus Mazes
Proof. It is clear that An pP q
the relation
P
0 when |P| ¡ n. We now prove that An respects
¸ ¹
P
|A|¤n
#A P p P
p
EA .
degA p
Calculate:
¸ ¹
An P
|A|¤n
#A P p P
p
EA degA p
¸ ¹
P
|A|¤n
#A P p P
p
degA p
¸
ä
rp : xÑysPS
|S|n
#S A
x
y
degP S ¹
p
1
degA S pPP degA p
degP S
# A P # S A
¸
¸
|A|¤n |S|n
The quantity
¹
P
p P
p
degA p
x ¹
rp : xÑysPS y
counts the number of ways to colour the passages of A (or EA ) in distinct
colours, with a selection of p colours available for (copies of) passage p. There
are exactly
degP S
degA S
distinct ways in which the passages of S can inherit the colours from A. Hence
the coeHcient of
¹
1
x
y
degP S
rp : xÑysPS
in the horrendous sum above is the number of ways to colour the passages of
S arbitrarily, with p colours to choose from for passage p. Consequently, the
sum equals
¸
¹
1
p
degP S
# S P
pPS
|S|n
¹
rp : xÑysPS
x y
¸
ä
rp : xÑysPS
|S|n
#S P
An pPq.
p
x
y
§2. Pure Mazes
69
Theorem 3.
The Ariadne functor factors through the quotient category Labyn , producing a functor
An : Labyn Ñ MSetn .
Proof. When P is pure,
An pa d P q ¸
ä
rxÑysPA
|A|n
a
#A P
a
¸
n
ä
rxÑysPA
|A|n
#A P
x
y
x
y
an An pPq.
There will be no more factorisation after this, for passing to the category
Labyn has the evect of making the Ariadne functor faithful. This is easy to see
from the theorem below, asserting that the pure mazes with exactly n passages
generate the category Labyn over Q bZ B. We shall even manufacture an inverse
of sorts to An .
§2. Pure Mazes
The Ariadne functor can be used to shed important light on the internal structure of the labyrinth categories.
Theorem 4.
The pure mazes are linearly independent in the category Laby.
Proof. Let Pm,j denote a pure maze of cardinality m, where all Pm,j are assumed
distinct. Suppose we have a relation
¸
¸
an,j Pn,j
j
an
1,j Pn 1,j
0
j
in LabypX, Y q, for some am,j P B. The nth Ariadne functor will kill all mazes
of cardinality greater than n, and the end result after application will be
¸
an,j An pPn,j q 0.
j
Since the Pn,j are distinct pure mazes, the An pPn,j q will all denote distinct multations. Hence all an,j 0. The claim now follows by induction.
Theorem 5.
The pure mazes constitute a basis for the category Labyn .
Proof. The proof for linear independence goes through exactly as before, because the Ariadne functor factors through Labyn . From the defining equations
for Labyn , we see that any maze will reduce to pure ones.
70
Chapter 4. Multi-Sets versus Mazes
Theorem 6.
The pure mazes with exactly n passages are linearly independent in the
category Labyn , and generate the category over Q bZ B.
Proof. Linear independence goes through as before. The defining equation for
Labyn can be written
an P
¸
a
P
P
# A P
|P| |A|¤n
a
A,
A
and since an Pa if |P | n, such a P may be expressed in terms of mazes with
more passages, provided division by integers is permissible.
As can be seen from the proof, it is in fact not necessary to have available
inverses for every integer. It suHces to invert the positive integers up to n.
Theorem 7.
The following categories are isomorphic:
B Labyn
n
B Laby
B bZ Z Labyn
B bZ Z Labyn .
Proof. The first equation is an immediate corollary of the pure mazes constituting a basis for Labyn . Let us prove the second one. By the theorem above,
Labyn is torsion-free. From the definition of Labyn ,
Q bZ B bZ Z Labyn .
n
n
B Laby will then embed in the former of those, and B bZ Z Laby in the latter.
Q bZ B Labyn
Because the Ariadne functor embeds Z Labyn in Z MSetn , which is free, it
follows that Z Labyn is free as well. By the isomorphism above, B Labyn will
then be free for arbitrary B, though it does not, in general, seem to possess a
preferred basis.
§3. The Theseus Functor
Example 1.
Let
P
21
1
1
1
/1
/ ,
/2
1
Q2
1
1
1
/1
/ 2 .
/
In Q bZ Laby3 , as may be checked, the following equation holds:
2It1,2u
1
2
1
1
/1
/2
P
Q.
Also, evidently PQ QP 0; hence the mazes 21 P and 21 Q would form a direct
sum system, splitting the set t1, 2u in two. But no such sets exist.
4
§3. The Theseus Functor
71
It is the purpose of this section to adjoin objects to the category Q bZ Labyn
so that some sets (like t1, 2u in the example) actually split as direct sums. This
will enable us to define an inverse to the Ariadne functor.
Lemma 1.
Let M be a torsion-free module, and let
ppxq P M b Brxs.
Then p 0 iff ppaq 0 for all integers a.
Proof. Induction on the degree of p.
Lemma 2.
In the category Q bZ Labyn , the following equation holds, for any set X:
IX
¸
|S|n
# S IX
1
S.
deg S
Proof. Use the previous lemma. Identify the coeHcient of an in the defining
equation for Labyn :
¸
a
an I X S.
S
# S P
|S|¤n
Evidently, the mazes
1
deg S S,
with S supported in IX , satisfy
1
1
ES E
deg S
deg T T
#
0
1
deg S S
if S
if S
T,
T.
Hence they form a direct sum system, though the objects themselves of the
system do not exist. There is a simple remedy for this: adjoin to the category
Q bZ Labyn an object
1
Im
S
deg S
for each such S.
However, we first observe that the category Q bZ Labyn is not the minimal
localisation of Labyn , for which this procedure makes sense. It may be verified
that the mazes
1
A,
deg A
where |A| n, and #A P for some pure and simple maze P, form a basis for
a subcategory of Q b Labyn which contains Labyn .
To this category we adjoin an object
Im
1
S
deg S
72
Chapter 4. Multi-Sets versus Mazes
for each multi-set S supported in some IX , and denote the resulting category
by
Laby`n .
The set X will now split into components:
X
à
Im
1
S.
deg S
|S|n
The category Laby`3 contains exactly five (isomorphism clas# S IX
Example 2.
ses of) objects. The set t1, 2u has been split up in two:
t1, 2u Im 21 P ` Im 21 Q.
4
Definition 2.
The nth Theseus functor
Tn : MSetn
Ñ Laby`n
is given by the following formulæ:
A ÞÑ Im
n ¹
ak
bk
k 1
1 ¤!
a
deg A aPA
ÞÑ
n !
¤
/ bk
1
ak
/a
1
)
)
.
k 1
It should be rather clear that this is indeed a (linear) functor, as composition in both categories is evectuated by “summing over all possibilities”. An
arbitrary multation transforms as
¤ !
1
a
deg µ
µ ÞÑ
Theorem 8.
pa,bqPµ
1
/b
)
.
There is an isomorphism of categories:
An
Laby`nh
'
MSetn
Tn
Proof. We first show that the Ariadne functor actually factors through the
category Laby`n . Let A be a multi-set with |A| n and #A P a pure maze.
Then
¹
ä
x
x
An pAq deg A ,
y
y
1 1
Ñy PA
x
x
Ñy PA
§4. The Category of Correspondences
73
and hence we may extend An by letting
ä
1
A ÞÑ
1 deg A
x
x
.
y
Ñy PA
Moreover, An maps the “virtual” biproduct system
IX
¸
|B|n
#B X
1 ¤!
b
deg B
/b
1
P
)
b B
in Labyn onto the “real” biproduct system
¸
|B|n
ιB
#B X
in MSetn , and we may consequently extend An to Laby`n by defining
An
1 ¤!
Im
b
deg B
/b
1
P
)
B.
b B
It is now easy to see that Tn and An are inverse to each other.
§4. The Category of Correspondences
For reference, we devote this last section to investigating the connection between the labyrinth category and the category of surjections explored by the
quartet Baues, Dreckmann, Franjou & Pirashvili in [1].
Let C be a category possessing weak pullbacks; that is, a finite number of
universal ways to complete an incomplete pullback square. For two objects
X, Y P C, a correspondence1 from X to Y is a diagram
Y o
/ X,
U
to be read from right to left. Suppose there exists a commutative diagram of
the following shape, with the middle column an isomorphism:
Y o
UO
/X
Y o
V
/X
We then identify the two correspondences
Y o
U
/X
Y o
V
/X .
1 Professor Pirashvili and his colleagues employ the word flèche, a rather unfortunate choice, as
this may also denote a single arrow.
74
Chapter 4. Multi-Sets versus Mazes
Construct a category Ĉ in the following way. Its objects are those of C.
Its arrows are formal sums of correspondences of C (identified under the just
described equivalence relation), in the free abelian group they generate. The
composition of correspondences is defined as:
Zo
V
/Y
Y o
/X
U
¸
Zo
W
/X
,
where the sum is taken over all weak pullbacks:
W
Z
}
V B
BB

B

!
Y
||
|~ |
U@
@@
@
X
(If C does indeed possess pullbacks, there will be no need to bother to these
formal sums, and composition can be defined simply as the pullback.) The
category Ĉ is pre-additive, and is called the category of correspondences in C.
It will now be observed that the category Sur of finite sets and surjections
possesses weak pullbacks. Namely, the condition
W
„ A P B tpa, bq P A B | αpaq βpbqu
(that the projections of W on A and B both be onto) is necessary and suHcient
for the following to be a weak pullback square:
W
/B
A
/P
α
β
We call A P B (the pullback in Set) the principal pullback.
y of surjection correspondences.
It is thus possible to build the category Sur
y n by forcing
We form a quotient category Sur
Y o
U
/X
0
whenever |U | ¡ n. This category, so vividly dissected in [1], turns out to be an
old acquaintance of ours.
Theorem 9.
yn
Sur
Proof. Define a functor
Z Labyn .
yn
Ξ : Sur
Ñ Z Labyn
§4. The Category of Correspondences
75
by ΞpX q X for any finite set X. The correspondence
ϕ
ϕ
Y o
U
ϕ
/X
y n will map to the pure maze X Ñ Y , of which the passages x
in Sur
number exactly
pϕ , ϕ q1 px, yq
(the cardinality of the fibre above px, yq P X Y ).
We now explain why this gives a functor. Let
ψ
Zo
ψ
V
ψ
/Y
,
ϕ
Y o
be two wedges, corresponding to the mazes Q : Y
ively. The number of passages x Ñ y in P equals
ϕ
U
ϕ
Ñ
y
/X
Ñ Z and P : X Ñ Y , respect-
|pϕ , ϕ q1 px, yq|,
and the number of passages y Ñ z in Q equals
|pψ , ψ q1 py, zq|.
The functoriality of Ξ follows from the observation that U Y V may be
naturally identified with Q b P, and subsets W „ U Y V with submazes
R „ Q b P.
Since the pure mazes form a basis, this functor is invertible.
Example 3.
An example will clarify the idea. The pure maze
co
za
zzzzz z
zzz
}zzzz
zz
z
}z
do
b
should be thought of as a correspondence
tc, du o tpc, aq, pd, aq1 , pd, aq2 , pd, bqu
where the maps are the obvious projections.
/ ta, bu ,
4
Curiously enough, even though
yn
Sur
Z Labyn
y and Laby are themselves not isomorphic. This
for all n, the categories Sur
Z
stems from the fact that Z Laby encodes functors from the category of free
y was built to encode functors from the category of
abelian groups, while Sur
free commutative monoids. These functor categories are not originally equivalent, but they will be, once polynomiality is assumed.
76
Chapter 4. Multi-Sets versus Mazes
Chapter 5
POLYNOMIAL
MAPS
And God said unto the animals: “Be fruitful and multiply.”
But the snake answered: “How could I? I am an adder!”1
Before embarking on the study of module functors, it seems worthwhile to gain
a reasonable understanding of module maps. The words of Professor Roby [19]
may serve as an opening vignette:
L’idée qui nous fit entreprendre le travail que nous présentons ici a son origine dans
le souci d’élargir l’arsenal des applications dont on dispose communément en algèbre. L’analyse classique dans les espaces Rn ou Cn utilise une grande diversité de
fonctions ou d’applications (continues, divérentiables, analytiques, holomorphes,
etc.) ; il en est de même, par exemple, dans l’étude des variétés pour lesquelles Rn
et Cn servent de modèle local. Mais dès qu’on considère des espaces vectoriels plus
généraux, et pis encore s’il s’agit de modules, on ne dispose plus guère que d’applications et de formes linéaires.
To us, a map of modules shall always denote an arbitrary map — in general non-linear. On those rare occasions when a linear map is actually under
consideration, we shall try to proclaim it a “homomorphism” quite loudly.
All modules will of course be B-modules, and our module maps will be
generalisations of ordinary polynomial maps (as defined on fields). The problem is then, naturally, how to form “polynomials” on modules, where no
multiplication is at hand.
One possible approach is to study maps that are (let us phrase it carefully) polynomial-like2 , in the sense that they satisfy certain equations somehow
thought to characterise polynomials. For example, a map ϕ : Ck Ñ Ck is
quadratic (polynomial of degree 2) iv, as may be checked, it satisfies the two
equations
ϕpx
y
zq ϕpx
yq ϕpy zq ϕpz xq
ϕpxq ϕpyq ϕpzq ϕp0q 0
1 In some retellings of this myth, it is said that God constructed a wooden table for the snakes
to crawl upon, since even adders can multiply on a log table. God is not assumed to be familiar
with tensor products.
2 På swensko: polynom-agtiga.
77
78
Chapter 5. Polynomial Maps
and
ϕpaxq a
ϕp0q
0
a
1
ϕpxq ϕp0q
a
2
ϕp2xq 2ϕpxq
ϕp0q ,
for any x, y, z P Ck and a P C. We observe the binomial coeHcients, which
explains why numerical rings will inevitably enter the theory.
It is perhaps surprising that the importance of the second condition was
not discovered until now. (We presume we are the very first to explore its
consequences.) The explanation for this could possibly be that (1) numerical
rings are not very much in use, and (2) people have only been interested in
maps of abelian groups, and never cared for the case of a general module. But
we digress.
The quantity
ϕpx y zq ϕpx y zq ϕpx yq ϕpy
ϕpxq ϕpyq ϕpzq ϕp0q
zq ϕpz
xq
is called the (second) deviation of ϕ. From the formula it should be clear how to
form higher-order deviations. In general, a map ϕ : M Ñ N of modules is said
to be numerical of degree n if it satisfies the two equations
ϕpx1 xn
and
ϕpaxq 1
q0
a
ϕpq
0
a
ϕpxq
1
a
ϕpx xq
2
,
for any a P B and x1 , . . . , xn 1 , x P M.
Apart from requiring a numerical base ring, such an approach has the
seeming disadvantage of producing maps that merely behave as polynomials,
without actually being polynomials. On the other hand, Roby’s strict polynomial
maps (he called them himself lois polynomes) actually look like polynomials. He
develops their theory in a journal article [19] spanning a hundred and thirty
pages, from which we quote:
[. . . ] la généralisation en vue devrait conduire à associer, à «quelque chose» qui
s’écrirait : x1 T1
xp Tp , une «autre chose» qui s’écrirait
q
¸
p
q
yi Qi T1 , . . . , Tp ,
i 1
les Qi étant cette fois des polynomes. Manifestement s’introduisent ici les modules
produits tensoriels [. . . ].
The definition Roby hints at is the following. A strict polynomial map
ϕ : M Ñ N is a natural transformation
ϕ: M b Ñ N
b
§1. Polynomiality
79
of functors CAlg Ñ Set. This may look even less like a polynomial than the
numerical maps defined above, but because of naturality, it is easily seen that
the following holds: for any u1 , . . . , uk P M there exist unique elements vX P N (only
finitely many of which are non-zero), X varying over multi-sets supported in rks, such
that
¸
ϕpu1 b x1 uk b xk q v X b xX ,
X
for all xj in all algebras.
It turns out that a similar property holds for numerical maps, namely: for
any u1 , . . . , uk P M there exist unique elements vX P N (only finitely many of which are
non-zero), X varying over multi-sets supported in rks, such that
ϕpx1 u1
xk u k q ¸
X
x
X
b vX ,
for all xj in all numerical algebras..
A remarkable fact then ensues, namely, that the numerical maps admit a
description similar in spirit to Roby’s framework. The resulting theorem is
that the map ϕ : M Ñ N is numerical (essentially) iv it can be extended to a
natural transformation
ϕ: M b Ñ N b of functors NAlg Ñ Set! This provides a beautiful and unexpected unification
of the two notions.
§1. Polynomiality
Let us begin by making an extremely general discussion of polynomiality, and
then identify the two notions we will actually make use of.
Let D „ Mod be a finitary algebraic category, by which is simply meant an
equational class in the sense of universal algebra. Since D is a subcategory of
Mod, the objects of D are first of all B-modules, possibly equipped with some
extra structure.
For a set of variables V , let
xV yD
denote the free algebra on V in D. That the free algebra exists is a basic fact
of universal algebra; see for example [4].
Definition 1.
Let M be a module, not necessarily in D. An element of
M b xx1 , . . . , xk yD
is called a D-polynomial over M in the variables x1 , . . . , xk .
A linear form over M in these variables is a polynomial of the form
¸
for some uj
P M.
u j b xj ,
80
Chapter 5. Polynomial Maps
Theorem 1: Ekedahl’s Esoteric Polynomiality Principle.
M and N be given, and a family of maps
ϕA : M b A Ñ N
b A,
Let two modules
A P D.
The following statements are equivalent:
A. For every D-polynomial ppxq ppx1 , . . . , xk q over M, there is a unique D-polynomial qpxq qpx1 , . . . , xk q over N, such that for all A P D and all aj P A,
ϕA pppaqq qpaq.
B. For every linear form lpxq over M, there is a unique D-polynomial qpxq over N,
such that for all A P D and all aj P A,
ϕA plpaqq qpaq.
C. The map
ϕ: M b Ñ N
b
is a natural transformation of functors D Ñ Set.
Proof. It is of course trivial that A implies B. Suppose statement B holds, and
consider a homomorphism χ : A Ñ B, along with finitely many elements uj P
M. Define
¸
lpxq u j b xj ,
and find the unique D-polynomial q satisfying B. Then, for any aj P A, there is
a commutative diagram of the following form, proving that ϕ is natural:
M bA
b
ϕA
/N
°
bA
b
1 χ
1 χ
M bB
ϕB
/N
bB
°
uj b aj
/ qpaq
uj b χpaj q
/ qpχpaqq
Thus, condition C holds.
Finally, suppose ϕ natural. We shall prove condition A. Given a D-polynomial
ppxq P M b xx1 , . . . , xk yD ,
define
For any A P D and aj
qpxq ϕxx1 ,...,x
y pppxqq .
k D
P A, define the homomorphism
χ : xx1 , . . . , xk yD Ñ A, xj ÞÑ aj .
§1. Polynomiality
81
Then since ϕ is natural, the following diagram commutes:
M b x x1 , . . . , xk y
b
x
ϕ
x1 ,...,xk
y
/N
b xx1 , . . . , xk y
1 χ
b
1 χ
M bA
/N
ϕA
bA
ppxq
/ qpxq
ppaq
/ qpaq
The uniqueness of q is evident, which proves A.
When the conditions of the theorem are fulfilled, we call ϕ
Definition 2.
a D-polynomial map from M to N.
According to part B of the theorem, ϕA maps
¸
u j b aj
ÞÑ qpaq,
for some (unique) D-polynomial q. This observation means that the Polynomiality Principle, in naïve language, amounts to the following. If we want the
coefficients aj (in some algebra) of the module elements uj to transform according to certain operations, the correct setting is the category of algebras using these same operations.
Example 1.
A Mod-polynomial map ϕ : M Ñ N is just
° a linear transforma°
tion M Ñ N. This is because, by B above, ϕB will map uj b rj to vj b rj for
all rj P B, and such a map is easily seen to be linear. Conversely, any module
homomorphism induces a natural transformation M b Ñ N b .
4
Example 2.
Let S be a B-algebra; then S Mod
map M Ñ N is a transformation
M bAÑN
„ Mod. An S Mod-polynomial
bA
which is natural in the S-module A. This is the same as a natural transformation
pM b Sq bS Ñ pN b Sq bS ,
which is an S Mod-polynomial map M b S Ñ N b S; or, as we noted in the
previous example, an S-linear map from M b S to N b S.
4
The two examples below will be the important ones.
Example 3.
A CAlg-polynomial map M Ñ N is a strict polynomial map, or
polynomial law,
in
the sense of [19]. Here condition B reads as follows.
For every
°
°
linear form uj b xj over M, there is a unique (ordinary) polynomial vX b xX over
N (X ranging over multi-sets), such that for all algebras A and all aj P A,
ϕA
¸
uj b aj
¸
vX
b aX .
Intuitively, the coeHcients of the elements uj “transform as ordinary polynomials”.
4
82
Chapter 5. Polynomial Maps
Example 4.
Suppose now that B is numerical, and consider the category
NAlg of numerical algebras over B. An NAlg-polynomial map M Ñ N is what
will be called
Forevery
° a numerical map. Condition B now reads as follows.
°
linear form uj b xj over M, there is a unique numerical polynomial vX b Xx over
N (X ranging over multi-sets), such that for all numerical algebras A and all aj P A,
ϕA
¸
uj b aj
¸
vX
b
a
.
X
Intuitively, the coeHcients of the elements uj “transform as numerical (binomial) polynomials”.
4
§2. Polynomial Maps
For historical reasons, we shall begin the exposition by defining the notion of
polynomial maps. Over general modules, this is much too weak a notion to be
useful. For example, it cannot be incorporated into Roby’s framework, which
is one indication it is faulty.
Presumably, it was Eilenberg and Mac Lane, who first studied non-additive
maps of abelian groups, introducing in [6] the so-called deviations of a map.
Let ϕ : M
Definition 3.
is the map
Ñ N be a map of modules. The nth deviation of ϕ
ϕpx1 xn
1
¸
q
I
of n
|I | ϕ
p1q
n 1
„rn 1s
¸
P
xi
i I
1 variables.
Let us, for clarity, point out that the diamond sign itself does not work as
an operator; the entity x y does not possess a life of its own, and cannot exist
outside the scope of an argument of a map.
It is an immediate consequence of the definition that
ϕpx1
xn
¸
1q I
„rn 1s
ϕ
♦ xi .
P
i I
Loosely speaking, the nth deviation measures how much ϕ deviates from
being polynomial of degree n. We have for example
ϕpx yq ϕpx yq ϕpxq ϕpyq
ϕpxq ϕpxq ϕp0q,
and, of course,
We abbreviate
ϕpq ϕp0q.
ϕ ♦x
n
ϕploooomoooon
x xq.
n
ϕp0q,
§3. Numerical Maps
83
Definition 4.
The map ϕ : M Ñ N is polynomial of degree n if its nth
deviation vanishes:
ϕpx1 xn 1 q 0
for any x1 , . . . , xn
1
P M.
This definition of polynomiality is the classical one for abelian groups.3
While this notion remains valid for arbitrary modules, it is clearly a poor
one, as it does not take the scalar multiplication into account. Recall that an
extra condition
ϕprxq rϕpxq
need be imposed on a group homomorphism to make it a module homomorphism (but that this is automatic when the base ring is Z!). In the next
section, we shall see what this equation generalises to.
§3. Numerical Maps
The base ring B of scalars will now be assumed numerical.
Definition 5.
The map ϕ : M Ñ N is numerical of degree (at most) n if it
satisfies the following two equations:
ϕpx1 xn
ϕprxq ņ
k 0
1
q 0,
x1 , . . . , xn
r
ϕ ♦x ,
k
k
r
1
PM
P B, x P M.
Observe that, when we speak of maps of degree n, we always mean degree
n or less.
It is easy to prove that, over the integers, the second equation above is
automatic, so that the concepts of polynomial and numerical map coincide.
Example 5.
ϕ is of degree 0 iv it is constant, for when n
equations read:
0 the above
ϕpx1 q ϕp0q ϕpx1 q 0,
ϕprxq r
ϕpq ϕp0q.
0
4
Example 6.
When n 1, the equations read as follows:
ϕpx1
x2 q ϕpx1 q ϕpx2 q
ϕp0q ϕpx1 x2 q 0,
3 Of course, Eilenberg and Mac Lane themselves do not deign to make this definition, but
instead move on to more important topics, like the computation of homology.
84
Chapter 5. Polynomial Maps
ϕprxq r
ϕpq
0
The map
r
ϕpxq ϕp0q
1
r pϕpxq ϕp0qq.
ψpxq ϕpxq ϕp0q
is then a homomorphism. Conversely, any translate of a homomorphism is of
degree 1.
4
Example 7.
Let B Z. It is a well-known fact, and not diHcult to prove,
that the numerical (polynomial) maps ϕ : Z Ñ Z of degree n are precisely the
ones given by numerical polynomials of degree n:
ϕpxq ņ
x
.
k
ck
k 0
4
Lemma 1.
For r in a numerical ring and natural numbers m
formula holds:
ņ
p1q
r
k
k
k m
k
m
p1q
n
r
m
¤ n, the following
rm1
.
nm
Proof. Induction on n. (The Numerical Transfer Principle is optional; the
induction will go through in any ring.)
Ñ N be polynomial of degree n. It is numerical (of
Theorem 2.
Let the map ϕ : M
degree n) iff it satisfies the equation
ϕprxq ņ
p1q r
m
n m
m 0
rm1
ϕpmxq,
nm
r
P B, x P M.
Proof. This follows from the lemma:
ņ
k 0
r
ϕ ♦x
k
k
ņ
ķ
k 0
ņ
m 0
ņ
r
k
p1q k m
m 0
p1qm
k m
p1q n m
m 0
ņ
k
ϕpmxq
m
p1qk
r
m
r
k
k
m
ϕpmxq
rm1
ϕpmxq.
nm
§4. The Augmentation Algebras
85
§4. The Augmentation Algebras
There is an algebraic way of describing numerical maps, which turns out to
be very fruitful. Recall that the free module on a set M is the set
BrM s !¸
aj rxj s aj
)
P B, xj P M
of formal (finite) linear combinations of elements of M. It obviously has a
module structure, and if M itself is a module (or even abelian group), it also
carries a multiplication, namely the sum multiplication
rxsrys rx ys,
It makes BrM s into a commutative, associative algebra
extended by linearity.
with unity r0s.
When M is not only a module, but an algebra, there is another natural
multiplication on BrM s, namely the product multiplication:
rxs rys rxys.
This multiplication has the identity element r1s, and is of course commutative
only if M is. This latter operation will make an apparition later on, in the
context of Morita equivalence. In the present discussion, we will assume M to
be a module only, and concentrate on the sum multiplication.
Let thus M be a module, and consider the map
M
Ñ BrM s,
x ÞÑ rxs.
We may form its nth deviation
Theorem 3.
px1 , . . . , xn 1 q ÞÑ rx1 . . . xn 1 s.
In the free algebra BrM s, the following formula holds:
rx1 . . . xn 1 s prx1 s r0sq prxn 1 s r0sq .
Proof. Simply calculate:
¸
r x1 . . . xn 1 s I
p1q
|I |
n 1
„rn 1s
¸
P
xi
i I
prx1 s r0sq prxn 1 s r0sq .
There is a filtration of BrM s, given by the decreasing sequence of ideals
In
p rx1 . . . xn 1 s
rrxs ņ
k 0
xi
PMq
r
k
♦x
k
r
P B, x P M
,
n ¥ 1.
86
Definition 6.
Chapter 5. Polynomial Maps
The nth augmentation algebra is the quotient algebra
BrM sn
BrM s{In .
If M is an algebra, then In is a two-sided ideal with respect to the product
multiplication, so the augmentation algebras will inherit this operation.
Theorem 4.
The map
δn : M
Ñ BrM sn ,
x ÞÑ rxs,
is the universal numerical map of degree n, in that every numerical map ϕ : M
degree n has a unique factorisation through it.
Ñ N of
/ BrM sn
ME
EE
EE
E
ϕ EEE
" N
δn
Proof. Given a map ϕ : M
Ñ N, extend it linearly to a homomorphism
ϕ : BrM s Ñ N.
The theorem amounts to the trivial observation that ϕ is numerical of degree
n iv it kills In .
The augmentation quotients of a free module M are given by the next theorem.
Theorem 5.
In the polynomial algebra Brt1 , . . . , tk s, let Jn be the ideal generated by
monomials of degree greater than n. Denote by pei qki1 the canonical basis of Bk . Then
ψ : Brt1 , . . . , tk s{Jn
tX
ÞÑ
♦ ei
P
i X
Ñ BrBk sn
pres r0sqX
is an isomorphism of algebras. In particular, BrBk sn is a free module.
Proof. The map
Brt1 , . . . , tk s Ñ BrBk sn
t
X
ÞÑ
♦ ei
P
i X
pres r0sqX
is clearly a homomorphism of algebras, and since it annihilates Jn , it factors
via Brt1 , . . . , tk s{Jn . This establishes the existence of ψ.
§4. The Augmentation Algebras
87
We now define the inverse of ψ. Each ti is nilpotent in Brt1 , . . . , tk s{Jn , and
so the powers
p1 ti qa
are defined for any a P B. Accordingly, for an element
x a1 e 1
P Bk ,
ak ek
we define
ra1 e1
χ : BrBk s Ñ Brt1 , . . . , tk s{Jn
ak ek s ÞÑ p1 t1 qa1 p1 tk qak
Jn .
We write this more succinctly as
rxs ÞÑ p1 tqx
Jn .
The map χ is linear by definition, and also multiplicative, since
χprxsrysq χprx
ysq p1
p1 tqx p1 tqy χpxqχpyq.
It maps In into Jn , because, when x1 , . . . , xn 1 P Bk ,
χprx1 xn
1
tqx
y
¸
sq J
„rn 1s
¸
J
1
P
xj j J
°
|J | p1 tq jPJ xj
p1q
n 1
„rn 1s
n
¹1
|J | χ p1qn
¸
p1 tqx 1 0.
j
j 1
Also, for r
χ
P B and x P Bk ,
rrxs ņ
m 0
r
m
♦x
m
χ rrxs p1 tqrx ppptq
1q
r
m
m 0
p1 tqrx r
m̧
ņ
m 0
ņ
r
m
m 0
ņ
j 0
m̧
ņ
m 0
p1qmj
p1qmj
j 0
r
m
m
j
rjxs
m
j
p1 tqjx
p1 tqx 1 m
r
pptqm ,
m
where, in the last step, we have let pptq p1 tqx 1. By the Binomial Theorem,
ppptq
1q
r
8̧
m 0
r
pptqm ,
m
88
Chapter 5. Polynomial Maps
but since the terms of index n 1 and higher yield an pn 1qst degree polynomial, the above diverence will belong to Jn . We therefore have an induced
map
χ : BrBk sn Ñ Brt1 , . . . , tk s{Jn ,
and it is easy to verify that ψ and χ are inverses to each other.
Example 8.
The isomorphism
ψ : BrB2 s2
is given by
Ñ Brt1 , t2 s{J2
rs ÞÑ 1
re1 s ÞÑ t1
re2 s ÞÑ t2
re1 e1 s ÞÑ t12
re1 e2 s ÞÑ t1 t2
re2 e2 s ÞÑ t22 .
4
§5. Properties of Numerical Maps
Let us elaborate somewhat on the behaviour of numerical maps, and investigate their elementary properties. To begin with, we note that the binomial
coeHcients themselves, considered as maps B Ñ B, are numerical. This is of
course hardly surprising, as they are given by polynomials in the enveloping
Q-algebra.
Theorem 6.
The binomial coefficient x ÞÑ
x
n
is numerical of degree n.
Proof. It is numerical of degree n in Z, and therefore also in B by the Numerical Transfer Principle.
Next, not only do the nth deviations of an nth degree map vanish, but its
lower order deviations are also quite pleasant.
Theorem 7.
holds:
The map ϕ : M
Ñ N is numerical of degree n iff the following equation
¸
ϕpa1 x1 ak xk q a
ϕ ♦x ,
X
X
rks
|X |¤n
#X
ai
Proof. If the equation is satisfied, it follows that
ϕpx1 xn
1
q
¸
1
ϕ ♦x
X
X
rn 1s
|X |¤n
#X
and
ϕpa1 x1 q ¸
r1s
|X |¤n
#X
a
ϕ ♦x
X
X
ņ
k 0
P B, xi P M.
0,
a1
ϕ ♦ x1 .
k
k
§5. Properties of Numerical Maps
89
Conversely, if ϕ is of degree n, a calculation in the augmentation algebra
BrM sn yields
ra1 x1 ak xk s pra1 x1 s r0sq prak xk s r0sq
8̧
8̧
q1 1
q1 1
a1
q1
♦ x1
q1
8̧
a1
q1
qk 1
8̧
qk 1
ak
qk
♦ xk
qk
♦ x1 ♦ xk .
ak
qk
qk
q1
The theorem follows after application of ϕ.
This proof is pure magic! It is absolutely vital that the calculation be
carried out in the augmentation algebra, as there would have been no way to
perform the above trick had the map ϕ been applied directly.
If ϕ is of degree 3, the following formulæ hold:
Example 9.
ϕpa1 x1 q a1
a1
a1
ϕpx1 q
ϕpx1 x1 q
ϕpx1 x1 x1 q
1
2
3
a1
a2
ϕpa1 x1 a2 x2 q ϕpx1 x2 q
1
1
a1
a2
a1
a2
ϕpx1 x1 x2 q
ϕpx1 x2 x2 q
2
1
1
2
a1
a2
a3
ϕpa1 x1 a2 x2 a3 x3 q ϕpx1 x2 x3 q.
1
1
1
4
We now have the following very explicit description of numerical maps.
Theorem 8.
The map ϕ : M Ñ N is numerical of degree n iff for any u1 , . . . , uk
P M there exist unique elements vX P N, X varying over multi-sets with #X „ rks and
|X | ¤ n, such that
¸ a
ϕpa1 u1 ak uk q vX ,
X
X
for any a1 , . . . , ak
P B.
Proof. Assume ϕ is numerical of degree n. By the preceding theorem, we have
ϕpa1 u1
¸
ak u k q I
„rks
¸
I
ϕ
♦ ai u i
P
i I
¸
„rks #X I
|X |¤n
a
ϕ
X
♦ ui
P
i X
90
Chapter 5. Polynomial Maps
¸
a
ϕ
X
„rks
|X |¤n
#X
♦ ui ,
P
i X
which establishes the existence of the elements vX .
We now prove uniqueness. Let Q „ rks be a multi-set, with qi
let
S tX | #X „ rks ^ @i : degS i ¤ qi u.
Then
ϕpq1 u1
q k uk q ¸
q
vX
X
X
¸
vQ
¸
P
X S
P zt u
X S Q
degQ i, and
q
vX
X
q
vX .
X
We see that vQ is determined by all vX , such that X precedes Q in the lexicographical ordering on the set of all multi-sets on rks, which can be identified
with Nk . By induction, each vX is uniquely determined.
Conversely, assume ϕ is of the form specified in the theorem. It then readily follows that
vX
ϕ
♦ ui
P
i X
for all X. In particular, the nth deviations of ϕ will vanish, and also
ϕpauq ņ
m 0
a
vm
m
ņ
m 0
a
ϕ ♦u .
m
m
And so, finally, we shall tie things together, and show that the definition
we have given of numerical map, “essentially” coincides with NAlg-polynomiality; this latter notion entailing a natural transformation
M bÑN
b
between functors NAlg Ñ Set. The subtle point here is that of degree. A
numerical map, as we have defined it, always comes with a degree, which is
of course not uniquely defined; it will also be numerical of any higher degree.
In particular, there is no such thing as a map that is simply numerical. This is
where the concept we have introduced divers from NAlg-polynomiality, for a
map could well be NAlg-polynomial of infinite degree.
Example 10.
Let U
ϕA : U
xu0 , u1 , u2 , . . .y be free on an infinite basis. The map
b A Ñ U b A,
¸
uk b ak
ÞÑ
¸
uk b
ak
k
is NAlg-polynomial, but not numerical of any finite degree n.
4
§6. Strict Polynomial Maps
91
In order to exclude such anomalies, we are obliged to introduce a boundedness condition. Let ϕ : M Ñ N be an NAlg-polynomial map. From the Polynomiality Principle, we know that for every linear form lpxq over M there is a
unique NAlg-polynomial qpxq over N, such that
ϕA plpaqq qpaq,
for all A P NAlg and all aj P A. We say that ϕ is of bounded degree n if the
degree of the polynomial q is uniformly bounded above by n (independent of
l).
Theorem 9.
The map ϕ : M Ñ N is numerical of degree n iff it may be extended to
a (unique) NAlg-polynomial map of bounded degree n.
Proof. If
ϕA : M b Ñ N
b
is an NAlg-polynomial map of bounded degree n, it is clear from the Polynomiality Principle that
ϕB : M Ñ N
has the property of Theorem 8.
Conversely, let a numerical map ϕ : M Ñ N be given. Given elements
u1 , . . . , uk P M, fix the elements vX from Theorem 8. We may then extend
ϕ in the obvious way to a natural transformation:
ϕA : M b A Ñ N
b A,
¸
uj b aj
ÞÑ
¸
vX
X
b
a
.
X
Some care is needed to ensure this map is well-defined on the tensor product,
but everything works out in the end, essentially because it is postulated to
work in the case A B.
§6. Strict Polynomial Maps
As we direct our attention toward strict polynomial maps, we no longer assume a numerical base ring.
Definition 7.
formation
A CAlg-polynomial map ϕ : M
M bÑN
Ñ N; that is, a natural trans-
b
between functors CAlg Ñ Set; which is of bounded degree n, will be called a
strict polynomial map of degree n.
It is clear that a strict polynomial map is also numerical of the same degree
(provided of course the base ring is numerical). If the base ring B is a Qalgebra, the two concepts coincide, for then every algebra is numerical.
Example 11.
A map is strict polynomial of degree 0 iv it is constant.
4
92
Chapter 5. Polynomial Maps
Example 12.
Let ψ : M
Ñ N be a homomorphism. Then
ψ b 1: M b Ñ N b defines a strict polynomial map of degree 1.
From this we see that in degrees 0 and 1, numerical and strict polynomial
maps always coincide.
4
Example 13.
Let B Z. The map
ξ : Z b Ñ Z b ,
1 b x ÞÑ 1 b
x
,
2
is numerical of degree 2, but not strict polynomial of any degree. The following tentative diagram, where β : t ÞÑ a, indicates the impossibility of defining
ξZrts .
Z b Zrts
b
1 β
ξZrts
ZbZ
/ Z b Zrts
ξZ
b
1bt
/?
1ba
/ 1b
1 β
/ ZbZ
a
2
Note that Zrts is not a numerical ring; there is no such thing as
t
2
!
4
Example 14.
Contrary to the situation for numerical maps, strict polynomial maps are not determined by the underlying maps. The most simple
example is probably the following. Let B Z, and define
ϕA : Z{2 b A Ñ Z{2 b A,
1 b x ÞÑ 1 b xpx 1q.
This is a non-trivial strict polynomial map of degree 2, and its underlying map
is zero!
We see here at play the well-known distinction between polynomials and
polynomial maps, the former class being richer than the latter. The point is
that the strict polynomial structure provides extra data, which makes the zero
map strict polynomial of degree 2 in a non-trivial way.
4
As mentioned earlier, strict polynomial maps were invented by Norbert
Roby, who designated them lois polynomes. The elementary facts enumerated
below, where we consider a strict polynomial map ϕ : M Ñ N, are all taken
from his article [19].
1. From the Polynomiality Principle, the following proposition is immediately deduced. For any u1 , . . . , uk P M there exist unique elements vX P N (only
finitely many of which are non-zero), X varying over multi-sets with X „ rks, such
that
¸
ϕpu1 b a1 uk b ak q v X b aX
X
§6. Strict Polynomial Maps
93
for all aj in all algebras.
We call vX the multi-deviation of ϕ of type X, with respect to the elements ui , and we shall denote this by the symbol
vX
ϕ ur s .
X
(This notation makes the subscript look like a divided power, which, in
fact, it is. See below.)
2. ϕ is of degree n iv ϕurX s
0 whenever |X | ¡ n.
3. ϕ is said to be homogeneous of degree n if
ϕpazq an ϕpzq
for all a in all algebras A and all z P M
ϕurX s 0 only when |X | n.
b A. This amounts to saying that
4. When ϕ is homogeneous of degree n, note that
ϕurns
ϕpuq.
5. Any ϕ will have a unique decomposition into homogeneous components, namely:
ϕpu1 b a1
u k b ak q 8̧
¸
| |n
ϕurX s
n 0 X
b aX .
There is some subtlety here. It is important to note that the above sum
is only locally finite. For a given
z u 1 b a1
u k b ak ,
the sum contains only a finite number of terms, but this number depends
on z, and, worse, need not be bounded above. Restricting attention to
strict polynomial maps of some given degree n, as we shall do, these
diHculties are avoided.
6. There is a fundamental relationship between homogeneous maps and
divided power algebras. For any module M there is a universal homogeneous map
γn : M
Ñ Γn pM q,
¸
ui b ai
ÞÑ
¸
|X |n
of degree n, through which every map ϕ : M
uniquely:
/ Γn pM q b A
M b AJ
JJ
JJ
JJ
ϕ JJJ
J% N bA
γn
°
urX s b aX
ÑN
of degree n factors
°
rX s X
/
ui b ai
|X |n u b a
OOO
OOO
OOO
OOO
'
°
|X |n ϕurX s b a
X
94
Chapter 5. Polynomial Maps
In other words, there is a canonical isomorphism between the module of
homogeneous polynomial maps of degree n from M to N and the module
of homomorphisms from Γn pM q to N.
7. Given ϕ, the map
Γn pM q Ñ N,
urX s
ÞÑ ϕur s
X
is a module homomorphism.
§7. The Divided Power Algebras
Divided power algebras are intimately connected to strict polynomial maps.
Roby, in all his prolificacy, has a great deal to say about them, and we once
again refer the reader to [19].
Like the free algebra, the divided power algebra ΓpM q comes equipped with
two natural multiplications.
Let first M be a module. We recall that the divided power multiplication on
ΓpM q is given simply by juxtaposition:
xr m s y r n s
xrms yrns .
The homogeneous components Γn pM q are not closed under this operation.
Let now M be an algebra. Then there is another multiplication on the nth
divided power module Γn pM q, which we now describe. Note that there is a
canonical map
Ñ Γn pM q b Γn pM q, px, yq ÞÑ xrns b yrns ,
which is universal for bihomogeneous maps of bidegree pn, nq out of M M.
δ: M M
Because the map
Ñ Γn pM b M q, px, yq ÞÑ px b yqrns ,
is bihomogeneous of degree pn, nq, it will have a unique factorisation through
ζ: M M
δ:
/ Γn pM q b Γn pM q
M MO
OOO
OOO
O
ζ OOO'
Γn pM b M q
δ
/ Γn pM q
Composition with the canonical (linear) map
px b yqrns ÞÑ pxyqrns ,
results in the following multiplication on Γn pM q:
Γn pM b M q Ñ Γn pM q,
Γn pM q b Γn pM q Ñ Γn pM q,
xrns b yrns
ÞÑ x y pxyqrns .
§7. The Divided Power Algebras
95
It will be called the product multiplication on Γn pM q.
Let now A and B be multi-sets of cardinality n, and let the variables xa and
yb be indexed over #A and #B, respectively. It is then easy to verify that the
following formula holds:
xrAs yrBs
¸
µ: A
ÑB
pxyqrµs
(where the sum is taken over multations, rather than sums of such). We do not
prove this formula here, as we shall establish a more general result in Chapter
7.
It deserves to be pointed out, and emphasised, that the nth divided power
module Γn pM q is not generated by the pure divided powers zrns , for z P M, as
the following example shows.
Consider in Γ3 pZ2 q a pure power
Example 15.
pa1 e1
a2 e2 qr3s
a31 e1r3s
r2s
r 2s
r2 s
a21 a2 e1 e2
a1 a22 e1 e2
3
r3 s
a2 e2 .
r2s
The coeHcients of e1 e2 and e1 e2 have the same parity (even or odd). Therer 2s
fore, it will be impossible to isolate e1 e2 as a linear combination of pure
powers.
4
Γn pM q is, however, “universally” generated by pure powers, in the following
sense.
Theorem 10: The Divided Power Lemma.
• A natural transformation
ζ : Γn pM q b Ñ N
b ,
between functors CAlg Ñ Mod, is uniquely determined by its effect on pure divided powers zrns (when z P M b A for some algebra A).
• More generally, a natural transformation
ζ : Γm pM q b Γn pM q b Ñ N
b
is uniquely determined by its effect on tensor products zrms b wrns of pure powers.
Proof. It suHces to show that if ζ vanishes on pure powers, it is identically zero.
Indeed, linear maps Γn pM q Ñ N correspond to homogeneous maps M Ñ N:
Γn pM q b A
O
ζ
/N
s9
s
s
s
s
γn
ss
ss
ss ζ
M bA
bA
°
u i b xi
O
rns
:/ 0
uu
u
u
uu
uu
u
uu
°
u i b xi
96
Chapter 5. Polynomial Maps
Since ζ ζγn 0, also ζ 0.
For the second part, proceed similarly, while noting that linear maps
Γm pM q b Γn pM q Ñ N
correspond to bihomogeneous maps
M `M
Ñ N.
Chapter 6
POLYNOMIAL
FUNCTORS
Och när jag stod där gripen, kall av skräck
och fylld av ängslan inför hennes tillstånd
begynte plötsligt mimans fonoglob
att tala till mig på den dialekt
ur högre avancerad tensorlära
som hon och jag till vardags brukar mest.
Harry Martinson, Aniara
With all preliminary work disposed of, we may finally begin our study of
polynomial functors proper. The first layer of bricks has been laid out with
minute detail, and will provide the rock solid foundations upon which to erect
our glorious temple. The chief aim of this monograph, it will be recalled, is to
study module functors. But, in order to describe these, it was necessary first
to gain a proper understanding of module maps. Only then could we hope for
some insight into the machinery of functors.
We now consider ourselves amply prepared for the task. Our incorporation of numerical and strict polynomial maps into the same framework (Professor Roby’s) will be seen to lead to a corresponding unification of the notions of numerical and strict polynomial functor. Here, again, can be seen the
deficiency of Professors Eilenberg and Mac Lane’s original notion, as functors
that are just polynomial do not allow for such a unified treatment.
The beginning of the chapter is comprised by a quick introduction to module functors in general. This is less of a luxurious indulgence than it may seem,
but rather an important reminder that our attention is restricted to a special class of functors, albeit rather large and general. Namely, we shall only
consider those module functors that are determined by their values on free and
finitely generated modules; this latter category admitting a smooth description,
and being pleasant in a multitude of ways. The question then arises as to what
functors possess this amiable property. We are by no means the first to study
these; from [3] we draw the characterisation of them as the right-exact functors
that commute with inductive limits.
Corresponding to the classes of numerical and strict polynomial maps are
the classes of numerical and strict polynomial functors, which we then set out
to explore. The chapter closes with a swift investigation of analytic functors,
97
98
Chapter 6. Polynomial Functors
which are characterised as the inductive limits of polynomial functors.
§1. Module Functors
Let
FMod
be the category of free modules, finitely or infinitely generated; and let
XMod
be the category of those modules that are finitely generated and free (the letter
X intended to suggest “eXtra nice modules”!). A module functor is a functor
XMod Ñ Mod
— and there should really be no need to point out that linearity will not be
assumed.
This may seem a meagre substitute for a functor defined on all modules,
but the restriction is not as heavy as one might think. As it turns out, a functor defined on the subcategory XMod has a canonical well-behaved extension
to the whole module category Mod. We now describe this extension process,
and thus convince ourselves (and hopefully the reader) that there is no serious imposition in considering only functors XMod Ñ Mod, as will be done
henceforth.
First, let us recall what it means for a functor, not necessarily additive, to
be right-exact.
Definition 1.
A functor F between abelian categories is right-exact if for
any exact sequence
A
α
/B
β
/C
/0
the associated sequence
F pA ` Bq
p q
p q
F α 1B
F 1B
/ F pBq
pq/
F pC q
F β
/0
is also exact.
This definition agrees with the usual one in the case of an additive functor.
In fact, the usual definition actually implies additivity of the functor, which
renders it useless for our purposes.
Theorem 1.
1. Any functor XMod Ñ Mod has a unique extension to a functor FMod
which commutes with inductive limits.
Ñ Mod
§1. Module Functors
99
2. Any functor FMod
Mod Ñ Mod.
Ñ
Mod has a unique extension to a right-exact functor
Proof. Since the result is well-known, we shall be content to sketch an outline
of its proof.
The first part follows from Lazard’s Theorem, which states that every flat
module is an inductive limit of finitely generated, free modules. A functor
G : XMod Ñ Mod may be extended to G : FMod Ñ Mod by putting
Gplim
ÝÑ Mα q lim
ÝÑ GpMα q,
for any inductive system pMα q of finitely generated, free modules. This definition is probably independent of the inductive system chosen.
The second part of the theorem is an immediate consequence of Theorem
2.14 in [3]. (The crucial point is that FMod is a subcategory of projective
generators that is closed under direct sums.) The extension procedure (which
essentially uses parts of the Dold–Puppe construction originally presented in
[5]) may be summarised as follows.
Let a functor F : FMod Ñ Mod be given. Given a module M, choose a free
resolution:
ψ
/M
/0
/P
Q
Let π and ξ denote the canonical projections:
Po
π
P`Q
ξ
/Q
Define the extension F : Mod Ñ Mod by the equation
F pM q F pP q
N
F pπq Ker F pπ
ψξq .
Again, this definition is independent of the particular free resolution chosen.
That F extends F is clear. When M is free, we may take the obvious free
resolution:
0 /
/M
/0
0
M
Here π 1M and ξ 0, and so
N
F pM q F pM q
N
F pM q
F pπq Ker F pπ
1F pM q pKer 1F pM q q
ψξq
F pM q
N
F p1M qpKer F p1M qq
F pM q{0 F pM q.
100
Chapter 6. Polynomial Functors
§2. Polynomial Functors
We now turn to interpreting our three notions of polynomiality in terms of
functors, in order from the weakest to the strongest.
Definition 2.
The functor F : XMod
degree (at most) n if every arrow map
Ñ Mod is said to be polynomial of
F : HompM, N q Ñ HompF pM q, F pN qq
is.
Example 1.
If F is polynomial of degree 0, then F pαq
homomorphism α. We have
1F pM q
F p0q for every
F p1M q F p0 : M Ñ M q F p0 : N Ñ M qF p0 : M Ñ N q,
and similarly
1F pN q
F p0 : M Ñ N qF p0 : N Ñ M q;
hence F pM q F pN q. The polynomial functors of degree 0 are thus the constant
ones.
4
If F is of degree 1, then
Example 2.
F pα
βq F pαq F pβq
F p0q 0,
or, equivalently,
F pα
The functor
βq F p0q pF pαq F p0qq
pF pβq F p0qq.
Epγq F pγq F p0q
is then additive.
Conversely, a functor F of the form
F pM q K ` EpM q,
with E additive, is polynomial of degree 1.
4
§3. Numerical Functors
In order to discuss numerical functors, we assume of course a numerical base
ring.
Definition 3.
The functor F : XMod
degree (at most) n if every arrow map
Ñ
Mod is said to be numerical of
F : HompM, N q Ñ HompF pM q, F pN qq
is.
§3. Numerical Functors
101
Note that, over the integers, the notions of polynomial and numerical
functor coincide.
Note also the inconspicuous assumption on uniformly bounded degree.
We shall presently see what happens when this assumption is dropped.
Example 3.
The functor F is numerical of degree 0 iv it is constant:
F pM q K.
4
Example 4.
The functor F is numerical of degree 1 iv it is of the form
F pM q K ` EpM q,
with E linear. Taking it on faith that there exist functors which are additive,
but not linear, we see that numericality is a stronger notion than polynomiality already in degree 1.
4
Example 5.
The most notorious examples of polynomial functors are no
doubt the classical algebraic functors: the tensor power T n pM q, the symmetric
power Sn pM q, the exterior power Λn pM q, and the divided power Γn pM q. They are
all numerical of degree n, because they commute with extension of scalars,
and are of bounded degree. For example, the map
TAn : A b T n pM q TAn pM q Ñ TAn pN q A b T n pN q
is clearly natural in all (numerical) algebras A.
A natural transformation η : F
homomorphisms
4
Ñ G of numerical functors is a family of
η pηM : F pM q Ñ GpM q | M
P XModq,
such that for any modules M and N, any numerical algebra A, and any
ω P A b HompM, N q,
the following diagram commutes:
A b F pM q
b /
A b GpM q
1 ηM
p q
p q
F ω
A b F pN q
(1)
G ω
b
1 ηN
/ A b GpN q
We shall denote the category of numerical functors of degree n by
Numn .
It is easy to see that it is abelian (the case B Z is well known). It is closed
under direct sums, and we will see in Chapter 9 that it possesses a small projective generator, so by the Morita Equivalence, it is in fact a module category.
102
Chapter 6. Polynomial Functors
Definition 4.
The numerical functor F is quasi-homogeneous of degree n
if the extension functor
F : Q bZ XMod Ñ Q bZ Mod
satisfies the equation
for any r
F prαq r n F pαq,
P Q bZ B and homomorphism α.
As remarked in connection with the category Labyn , demanding
F prαq r n F pαq
just for r P B would be insuHcient to yield interesting results. Being quasihomogeneous is, in a sense, the closest a numerical functor can come to being
homogeneous.
The category of quasi-homogeneous functors of degree n will be denoted
by
QHomn .
§4. Properties of Numerical Functors
Let us hasten to point out that our definition of natural transformation is
unnecessarily complicated. A consequence of Theorem 8 of Chapter 5 is that a
polynomial functor is uniquely determined by its underlying functor. In view
of this, the following theorem is hardly surprising. The reason for adopting
the more complicated condition as definition, is to conform to the situation
for strict polynomial functors.
Theorem 2.
η : F Ñ G.
The diagram (1) commutes for any (ordinary) natural transformation
Proof. Consider homomorphisms
α1 , . . . , αk : M
Ñ N.
Assume that
F pa1 b α1
Gpa1 b α1
ak b αk q ¸
ak b αk q ¸
µ
ν
a1
m1
a1
n1
ak
mk
ak
nk
b βµ
b γν ,
for any a1 , . . . , ak in any numerical algebra A, where we have abbreviated
µ pm1 , . . . , mk q and ν pn1 , . . . , nk q.
§4. Properties of Numerical Functors
103
The naturality of η ensures that
¸
a1
m1
µ
¸
m1
a1
ηN βpm1 ,0,... q
m1
Successively putting a1
¸
a1
n1
ν
ak
γν ηM .
nk
a3 0, to obtain
Specialise first to the case a2
ak
ηN βµ
mk
¸
a1
γ
ηM .
n1 pn1 ,0,... q
n1
0, 1, 2, . . . leads to
ηN βpm ,0,... q γpm ,0,... q ηM
1
1
for all m1 . Proceeding inductively, one shows that
γµ ηM
ηN βµ
for all µ. The commutativity of the diagram (1), for
ω a1 b α1
ak b αk ,
is then demonstrated by the following instantiation:
/ b b ηM pxq
bbx
°
a1
µ m1
a
k
mk
°
b b β µ p xq
/
ma b b ηN βµ pxq
ma b b γµ ηM pxq
a1
µ m1 °
a1
µ m1
k
k
k
k
The subsequent theorem is the first to illustrate a recurring theme: that a
numerical functor of degree n is essentially determined by its action on Bn .
Theorem 3.
degree n.
The following conditions are equivalent on a polynomial functor F of
A.
F prαq ņ
k 0
r
F ♦α ,
k
k
for any scalar r and homomorphism α (numerical functor).
B.
F prαq ņ
p1q n m
m 0
for any scalar r and homomorphism α.
r
m
rm1
F pmαq,
nm
104
Chapter 6. Polynomial Functors
A1 .
ņ
F pr 1Bn q r
F ♦ 1Bn
k
k
k 0
,
for any scalar r.
B1 .
F pr 1Bn q ņ
m 0
p1qnm
rm1
F pm 1Bn q,
nm
r
m
for any scalar r.
Proof. That A and B are equivalent follows from Theorem 2 of Chapter 5, as
does the equivalence of A1 and B1 . Clearly B implies B1 , so there remains to
establish that B1 implies B.
Hence assume B1 , and put
p1qnm
Zm
r
m
In the case q ¤ n the equation
F pr 1Bq q ņ
m 0
rm1
.
nm
Zm F pm 1Bq q
holds, because 1Bq factors through 1Bn .
Consider now the case q ¡ n. By induction, assume the formula holds for
q 1. Letting πi as usual denote canonical projections, we calculate:
F pr 1Bq q F prπ1
¸
I
I
€rqs
p1q |I | F
ņ
m 0
¸
P
rπi
i I
p1q |I |
¸
ņ
Zm
I
€rqs
ņ
q
€rqs
m 0
rπq q
q
¸
Zm F
m 0
p1q |I | F
Zm F pmπ1
q
¸
P
mπi
i I
¸
P
mπi
i I
mπq q ņ
m 0
Zm F pm 1Bq q.
The third and sixth steps are because the qth deviation vanishes. This shows
that the equation holds for 1Bq , for any q.
Finally, in the case of an arbitrary homomorphism α : Bp Ñ Bq , we have
F prαq F pr 1Bq qF pαq
§5. The Hierarchy of Numerical Functors
ņ
m 0
105
Zm F pm 1Bq qF pαq ņ
m 0
Zm F pmαq,
and the proof is finished.
The following very pleasant formula is an immediate consequence of the
corresponding formula for maps.
Theorem 4.
The module functor F is numerical of degree n iff for any scalars ai and
homomorphisms αi , the following equation holds:
F pa1 α1 ak αk q ¸
rks
|X |¤n
#X
a
F ♦α .
X
X
Proof. Theorem 7 of Chapter 5.
Example 6.
If F is of degree 3, the following formulæ hold:
F pa1 α1 q a1
a1
F pα1 q
F pα1 α1 q
1
2
a1
F pα1 α1 α1 q
3
a1
a2
a1
a2
F pa1 α1 a2 α2 q ϕpα1 α2 q
F pα1 α1 α2 q
1
1
2
1
a1
a2
F pα1 α2 α2 q
1
2
a1
a2
a3
F pa1 α1 a2 α2 a3 α3 q F pα1 α2 α3 q.
1
1
1
4
§5. The Hierarchy of Numerical Functors
We say that a map ϕ, or a family of such, is multiplicative if
ϕpzqϕpwq ϕpzwq,
whenever z and w are entities («quelques choses») such that the equation
makes sense, and also
ϕp1q 1,
where the symbol 1 is to be interpreted in a natural way (usually diverently on
each side). An ordinary functor is the prime example of such a multiplicative
family.
Also, we say that a family of maps is of (uniformly) bounded degree, if
every map in the family is numerical of some fixed degree n.
106
Chapter 6. Polynomial Functors
Theorem 5.
algebras:
Consider the following constructs, where A ranges over all numerical
A. A family of ordinary functors EA : A XMod Ñ A Mod, commuting with extension
of scalars.
B. A functor J : XMod Ñ Mod, with arrow maps
JA : HomA pA b M, A b N q Ñ HomA pA b J pM q, A b J pN qq
that are multiplicative and natural in A.
C. A functor F : XMod Ñ Mod, with arrow maps
FA : A b HomB pM, N q Ñ A b HomB pF pM q, F pN qq
that are multiplicative and natural in A (numerical maps).
Constructs A and B are equivalent, but weaker than C. If, in addition, the arrow maps
are assumed to have uniformly bounded degree, all three are equivalent.
Proof. Given E, define J by
J pM q EB pM q
and the diagram:
HomA pA b M, A b N q
/ HomA pEA pA b M q, EA pA b N qq
O
EA
J
+
HomA pA b EB pM q, A b EB pN qq
Conversely, given J, define the functors E by the equations
EA pM q A b J pM q
and
HomA pA b M, A b N q
/
HomA pA b J pM q, A b J pN qq.
EA J
Also, it is easy to define J from F; simply let
J pM q F pM q,
and use the following diagram:
A b HomB pM, N q
O
HomA pA b M, A b N q
F
J
/ A b HomB pF pM q, F pN qq
/ HomA pA b F pM q, A b F pN qq
§5. The Hierarchy of Numerical Functors
107
The left column in the diagram is an isomorphism as long as M and N are free.
The diHcult part is defining F from J, provided that J is indeed of bounded
degree n. The following proof is modelled on the corresponding proof for
strict polynomial functors in [20]. Let M and N be two modules, and let A be
any numerical algebra. Find a free resolution
/ J pM q
/ Bpκq
Bpλq
/0,
and apply the contra-variant, left-exact functor
HomA pA b , A b J pN qq,
to obtain a commutative diagram:
0
/ HomA pA b J pM q, A b J pN qq
O
/ pA b J pN qqκ
O
ι
σ
/ pA b J pN qqλ
ζ
J
A b HompM, N q
δn
/ A b BrHompM, N qsn
The homomorphism
ιJ : A b HompM, N q Ñ pA b J pN qqκ
may be split up into components
pιJ qk : A b HompM, N q Ñ A b J pN q,
for each k P κ. Those are numerical of degree n, and will factor over δn via
some linear ζk . Together they yield a linear map
ζ : A b BrHompM, N qsn
making the above square commute.
Now, σζδn σιJ 0, which gives σζ
row, ζ factors via some homomorphism
ξ : BrHompM, N qsn
Because
Ñ pA b J pN qqκ ,
0.
By the exactness of the upper
Ñ HompJ pM q, J pN qq.
ζδn ιξδn
and ι is one-to-one, we also have J ξδn . The following diagram will therefore
commute:
ιJ
HompJ pM q, J pN qq
O
i
J
HompM, N q
ι
ξ
δn
/ J pN qpκq
O
ζ
/ BrHompM, N qsn
Since J factors over BrHompM, N qsn , it is numerical of degree n, and so may be
used to construct F.
108
Chapter 6. Polynomial Functors
We thus obtain the following hierarchy of functors.
• Numerical functors, as defined previously, have bounded degree, and satisfy all three conditions A, B and C.
• A functor satisfying condition C, but with no assumption on the degree,
will be called locally numerical.
• A functor satisfying the weaker conditions A and B, again without any
assumption on the degree, will be called analytic.
Example 7.
The classical algebraic functors T , S, Λ and Γ are all analytic,
for they evidently satisfy condition A of the theorem.
Of these, only Λ is locally numerical. This is because, when n ¡ p, the
module
Λn pBp q 0,
and hence, for given p and q, the map
Λ : HompBp , Bq q Ñ HompΛpBp q, ΛpBq qq
is numerical of degree maxpp, qq.
4
§6. Strict Polynomial Functors
The base ring B is now no longer assumed numerical.
Definition 5.
The functor F : XMod Ñ Mod is said to be strict polynomial
of degree n if the arrow maps
F : HompM, N q Ñ HompF pM q, F pN qq
have been given a (multiplicative) strict polynomial structure.
A strict polynomial functor is also numerical of the same degree, provided
the base ring is numerical. If the base ring B is a Q-algebra, the two concepts
coincide, for then every algebra is numerical.
We remark that, comparable to the situation for maps, strict polynomial
functors are not determined by their underlying functors.
Example 8.
Because numerical and strict polynomial maps coincide in degrees 0 and 1, the same holds true for functors.
4
Example 9.
The classical algebraic functors T n , Sn , Λn and Γn are in fact
strict polynomial of degree n, because they are natural in all algebras, not just
numerical ones.
4
§6. Strict Polynomial Functors
109
Example 10.
The following singular example may serve as a warning. Let
B be a numerical ring, let p be a prime, and consider the functor
F pM q Sp pM q{pSp pM q.
F inherits from Sp the property of being homogeneous of degree p. On the
other hand, because p | r p r by Fermat’s Little Theorem, the underlying
functor of F takes
F prαq Sp prαq r p Sp pαq rSp pαq rF pαq.
Moreover, by the Binomial Theorem,
pa
so
F pα
bqp
βq Sp pα
ap
bp mod p,
βq Sp pαq
Sp pβq F pαq
F pβq.
The underlying functor of F is linear!
Quite obviously, F may also be given the structure of homogeneous functor
of degree 1, furnishing us with two diverent strict polynomial structures on
the same functor, and even of diverent degrees.
4
By a natural transformation η : F
mean a family of homomorphisms
Ñ G of strict polynomial functors, we
η pηM : F pM q Ñ GpM q | M
P XModq,
such that for any modules M and N, any algebra A, and any
ω P A b HompM, N q,
the following diagram commutes:
A b F pM q
b /
A b GpM q
1 ηM
p q
p q
F ω
A b F pN q
G ω
b
1 ηN
/ A b GpN q
We shall denote by
SPoln
the category of strict polynomial functors of degree n. It is well known to
be abelian, and, like the category of numerical functors, it is in fact a module
category. However, rather than consider arbitrary strict polynomial functors,
we shall usually limit our attention to homogeneous ones, as any strict polynomial functor decomposes as a direct sum of such.
110
Chapter 6. Polynomial Functors
§7. The Hierarchy of Strict Polynomial Functors
As in the numerical case, we have the following three equivalent characterisations of strict polynomial functors. The words (uniformly) bounded degree will
here be taken to mean “strict polynomial of bounded degree”.
Theorem 6.
Consider the following constructs, where A ranges over all algebras:
A. A family of ordinary functors EA : A XMod Ñ A Mod, commuting with extension
of scalars.
B. A functor J : XMod Ñ Mod with arrow maps
JA : HomA pA b M, A b N q Ñ HomA pA b J pM q, A b J pN qq
that are multiplicative and natural in A.
C. A functor F : XMod Ñ Mod with arrow maps
FA : A b HomB pM, N q Ñ A b HomB pF pM q, F pN qq
that are multiplicative and natural in A (strict polynomial maps).
Constructs A and B are equivalent, but weaker than C. If, in addition, the arrow maps
are assumed to have uniformly bounded degree, all three are equivalent.
Proof. The proof is exactly analogous to the one given for polynomial functors,
except that, in the proof that B implies C, the module
n
à
Γk Hom M, N
p
q
k 0
is used in place of BrHompM, N qsn . The details are found in [20].
As in the numerical case, we obtain the following hierarchy.
• Strict polynomial functors, as defined previously, have bounded degree, and
satisfy all three conditions A, B and C.
• A functor satisfying condition C, but with no assumption on the degree,
will be called locally strict polynomial.
• A functor satisfying the weaker conditions A and B, again without any
assumption on the degree, will be called strict analytic.
Example 11.
The functors T , S, and Γ are in fact strict analytic, and Λ is
locally strict polynomial.
4
§8. Homogeneous Functors
111
§8. Homogeneous Functors
There are some superficial similarities between numerical and strict polynomial functors — the first and most glaringly obvious one being their respective
definitions. Parts of their theories indeed run exactly in parallel. Therefore, it
may come as a surprise that, in fact, some very fundamental diverences exist
between the two genera.
There is especially one property of strict polynomial functors that is lacking for numerical ones, and that is the ability to split into a sum of homogeneous components. We saw this happen already in the case of maps.
Definition 6.
The functor F : XMod Ñ Mod is said to be homogeneous of
degree n if the arrow maps
F : HompM, N q Ñ HompF pM q, F pN qq
have been given a (multiplicative) homogeneous structure.
The category of homogeneous functors will be denoted by
Homn .
We shall always prefer this category over SPoln , since, according to the following theorem, nothing essential will be lost by considering homogeneous
functors only.
Theorem 7.
A strict polynomial functor decomposes as a unique direct sum of homogeneous functors. The only possible natural transformation between homogeneous functors of different degrees is the zero transformation. Consequently,
SPoln
n
à
Homk .
k 0
Proof. See [20].
Example 12.
If Fk is homogeneous of degree k, the direct sum
n
à
Fk
k 0
will provide a strict polynomial functor of degree n. By the theorem, this is
the generic situation. On the other hand, a numerical functor is in general
“more than the sum of its parts”. An example of this phenomenon cannot be
given at this stage, since we do not yet know of a functor that is numerical,
but not strict polynomial.
4
112
Chapter 6. Polynomial Functors
§9. Analytic Functors
We now examine the analytic and strict analytic functors. They will come into
play later, when we consider operads.
Theorem 8.
The strict analytic functors are precisely the direct sums (or, equivalently, inductive limits) of strict polynomial functors.
Proof. See [20].
Example 13.
The strict analytic functors T , S, Λ, and Γ all decompose as
infinite direct sums of homogeneous functors. This is the generic situation. 4
Lemma 1.
Let F be an analytic functor, let u P F pP q, and define the subfunctor G by
GpM q xF pαqpuq α : P
Ñ My .
Consider the natural transformation
ξ : HompP, q Ñ F,
given by
ξN : HompP, N q Ñ F pN q
α ÞÑ F pαqpuq.
If ξN is numerical of degree n, then so is
G : pM, N q Ñ HompGpM q, GpN qq
for all M. In particular:
• If all ξN are numerical, then G is locally numerical.
• If all ξN are numerical of bounded degree, then G is numerical.
Proof. Observe that the modules GpM q are invariant under the action of F.
Thus, G is indeed a subfunctor of F.
Suppose ξN is numerical of degree n. Then, for all homomorphisms
α, αi : P
ÑN
and scalars r, the following equations hold:
#
F pα1 αn 1 qpuq 0
°n
F prαqpuq m0 mr F p♦m αq puq.
This implies that, for all homomorphisms
β, βi : M
Ñ N,
γ: P
Ñ M,
§9. Analytic Functors
113
and scalars r, the following equations hold:
#
Hence
F pβ1 βn 1 qF pγqpuq 0
°n
F prβqF pγqpuq m0 mr F p♦m βq F pγqpuq.
F pβ1 βn
and
F prβq on GpM q, which means that every
ņ
m 0
1
q0
r
F ♦β
m
m
G : pM, N q Ñ HompGpM q, GpN qq
is indeed numerical of degree n.
Theorem 9.
functors.
The analytic functors are precisely the inductive limits of numerical
Proof. Step 1: Inductive limits of numerical, or even analytic, functors are analytic. Let
the functors Fi , for i P I, be analytic. For any
α P HomA pA b M, A b N q,
we have
Therefore
Fi pαq : A b Fi pM q Ñ A b Fi pN q.
lim
ÝÑ Fi pαq : A b lim
ÝÑ Fi pM q Ñ A b lim
ÝÑ Fi pN q,
since tensor products commute with inductive limits, which yields a map
lim
ÝÑ Fi : HomA pA b M, A b N q Ñ HomA pA b lim
ÝÑ Fi pM q, A b lim
ÝÑ Fi pN qq,
establishing that lim
ÝÑ Fi is analytic.
Step 2: Analytic functors are inductive limits of locally numerical functors. Let F be
analytic. The maps
F : HomA pA b M, A b N q Ñ HomA pA b F pM q, A b F pN qq
are then multiplicative and natural in A. To show F is the inductive limit of
locally numerical functors, it is suHcient to construct, for any given module
P and element u P F pP q, a locally numerical subfunctor G of F, such that
u P GpP q.
To this end, define G as in the lemma:
GpM q xF pαqpuq α : P
Ñ My .
114
Clearly u
that
Chapter 6. Polynomial Functors
P GpPq.
By the lemma, G is locally numerical, if only we can show
ξM : HompP, M q Ñ F pM q
is always numerical (of possibly unbounded degree).
We make use of the following fact. The module HompP, M q is finitely
generated and free, because P and M are. Let ε1 , . . . , εk be a basis.
Let s1 , . . . , sk be free variables, and let
AB
Since
F
¸
si b εi
s1 , . . . , sk
.
P HomA pA b F pPq, A b F pM qq,
we may write
F
¸
si b εi
p1A b uq ¸
|X |¤n
s
X
b vX P A b F pM q
for some n, and hence
ξM
¸
si εi
F
¸
si εi
puq ¸
s
vX .
X
|X |¤n
Since the εi generate HompP, M q, it follows that ξM is numerical of degree n.
Step 3: Locally numerical functors are inductive limits of numerical functors. Let F
be locally numerical, and, given P and u P F pP q, define G and ξ as before. We
shall show that G is numerical by showing that ξ is numerical of some fixed
degree.
Let αi : P Ñ M be homomorphisms, let
BB
s1 , . . . , sk
,
C
B
s1 , . . . , sk , t
be free numerical rings, and consider the algebra homomorphism
τ : B Ñ C,
si
ÞÑ tsi .
There is a commutative diagram:
B
τ
C
B b HompP, M q
F
b
b
τ 1
C b HompP, M q
/ B b HompF pP q, F pM qq
τ 1
F
/ C b HompF pP q, F pM qq
§9. Analytic Functors
115
As a consequence, we obtain, for any homomorphisms αi : P
°
°
Consider now
F: B
s1 , . . . , sk
/ F p° si b αi q
si b αi
tsi b αi
/
pτ b 1qF°p° si b αi q
F p tsi b αi q
b HompP, M q Ñ B
and write
F
¸
Ñ M:
si b αi
s1 , . . . , sk
b HompF pPq, F pM qq,
¸
s
X
X
b βX ,
for some homomorphisms βX : F pP q Ñ F pM q.
Similarly, from contemplating
F: B
t
b HompP, Pq Ñ B b HompF pPq, F pPqq,
t
we may write
F pt b 1P q ¸
b γm ,
t
m
¤
m n
for some number n and homomorphisms γm : F pP q
fixed, and only depends on F.
We now have
¸
X
ts
X
b βX pτ b 1q
¸
X
¸
s
X
b βX
Ñ F pPq. Observe that n is
pτ b 1qF si b αi
¸
F tsi b αi
¸
F si b αi F pt b 1P q
¸
X
s
X
¸ ¸
¤
X m n
b βX
s
X
t
m
¸
¤
m n
t
m
b γm
b β X γm .
The right-hand side, and therefore also the left-hand side, is of degree n in t,
whence βX 0 when |X | ¡ n.
116
Chapter 6. Polynomial Functors
Consequently,
ξM
¸
si αi
F
and ξ is numerical of degree n.
¸
si αi
p uq ¸
|X |¤n
s
βX puq,
X
Chapter 7
DEVIATIONS
AND
CROSS-EFFECTS
[. . . ] je donnerais bien cent sous au mathématicien qui me démontrerait par une
équation algébrique l’existence de l’enfer.
Honoré de Balzac, La Peau de chagrin
We have now arrived at the arguably most technical chapter of the monograph, being virtually nothing more than a collection of formulæ. These will
be established with the pronounced goal of examining the internal structure
of functors, essentially “taking them apart to see what makes them tick”. Mechanics employ a technical term for this process: reverse engineering.
The central concept in the theory of polynomial functors has hitherto been
that of cross-effects. Ever since they were first designed by Professors Eilenberg
and Mac Lane in 1954 ([6]), they have been the algebraist’s vivisection tool par
préférence, being to module functors what the deviations are to module maps.
Recall that the first deviation measures how much a module map ϕ deviates
from being aHne, as testified by the equation
ϕpx
yq ϕpq
ϕpxq
ϕpyq
ϕpx yq.
A corresponding equation for functors then ensues, namely the direct sum
decomposition
F pX ` Y q F : pq ` F : pX q ` F : pY q ` F : pX | Y q,
valid for any module functor F, and any modules X and Y . The quantities
ϕpq and F : pq are the “constant terms”. Disregarding these, the first crossevect F : pX | Y q, like the first deviation ϕpx yq, measures the deviation from
additivity. Indeed, polynomial functors of degree n may be defined, as did
originally Eilenberg and Mac Lane ([6]), as those with vanishing nth crossevects.
But the cross-evects carry an inherent deficiency, which greatly reduces
their utility, in that they do not take scalar multiplication into account. Of
course, neither did the deviations, but there we were able to save the day
117
118
Chapter 7. Deviations and Cross-Evects
by inventing a second equation involving binomial coeHcients, marking the
transition from polynomial to numerical maps. This flexibility is lost when
passing to cross-evects, which would seem to render them somewhat obsolete.
Nevertheless, we shall later see how to put them to good use.
§1. The Deviations
We first make a more detailed study of deviations in the context of functors.
We shall obtain formulæ for the composition of deviations, the deviations of
a tensor product, and the deviations of a composition. The results are valid
for any module functors F and G over any base ring B.
The symbol
M „X Y
shall denote that M is a subset of X Y , and that both the canonical projections
are onto.
Lemma 1.
Let m and n be natural numbers, and let L „ rms rns. Then
¸
„ „rmsrns
p1q|K| 0,
L K
unless L is of the form P Q, for some P
„ rms, Q „ rns.
Proof. If L is not of the given form, there exists an pa, bq, which is not in L,
but such that some pa, jq and some pi, bq are in L. Then, for any set K „
rms rns containing pa, bq, K will satisfy the given set inclusions iv K ztpa, bqu
does. Because the cardinalities of these sets diver by 1, the corresponding
terms in the above sum will have opposite signs, and hence cancel.
Lemma 2.
Let m, n, p, and q be natural numbers. Then
¸
rpsrqs„K „rmsrns
p1q|K| p1qm
n p q pq
.
Proof. Let Y pm, n, kq denote the number of sets K of cardinality k satisfying
rps rqs „ K „ rms rns.
The formula is evidently true for m p and n q, for then Y pp, q, pqq 1, and
all other Y pp, q, kq 0.
We now do recursion. Consider the pair pm, nq P rms rns. The sets K
containing pm, nq will fall into two classes: those where pm, nq is mandatory in
order to satisfy K „ rmsrns, and those where it is not. For the latter class we
may proceed as in the preceding proof. Taking such a K and removing pm, nq
will yield another set counted in the sum above, but of cardinality decreased
by 1. Since these two types of sets exactly pair ov, with opposing signs, their
contribution to the given sum is 0.
Consider then those K of which pm, nq is a mandatory element. They fall
into three categories:
§1. The Deviations
119
• Some pm, jq P K, for 1 ¤ j ¤ n 1, but no pi, nq P K, for 1 ¤ i ¤ m 1. The
number of such sets is Y pm, n 1, k 1q.
• No pm, jq P K, for 1 ¤ j ¤ n 1, but some pi, nq
The number of such sets is Y pm 1, n, k 1q.
• No pm, jq P K, for 1 ¤ j ¤ n 1, and no pi, nq
number of such sets is Y pm 1, n 1, k 1q.
P K, for 1 ¤ i ¤ m 1.
P K, for 1 ¤ i ¤ m 1. The
Induction yields
¸
p1qk Y pm, n, kq
k
¸
p1qk Y pm, n 1, k 1q
k
p1qm n1 p q
p1qm n p q pq ,
Y pm 1, n, k 1q
p1qm1
pq
Y pm 1, n 1, k 1q
p1qm1
n p q pq
n 1 p q pq
as desired.
Theorem 1: The Deviation Formula.
¸
F pα1 αm q F pβ1 βn q K
„rmsrns
♦ αi βj
F
pi,jqPK
.
Proof. We have
¸
K
„rmsrns
♦ αi βj
F
pi,jqPK
¸
L
L
K
¸
„rmsrns
p1q|L| F ¸
„rmsrns L„K „rmsrns
¸
¸
„rmsrns
„rmsrns L„K
p1q|K||L| F P
„rms
¸
pi,jqPL
p1q|P||Q| F P Q
¸
p1qm|P| F
¸
P
i P
p1q|K||L| F αi βj pi,jqPPQ
αi
¸
Q
„rns
pi,jqPL
pi,jqPL
αi βj p1q|K|
αi βj p1qm
p1qn|Q| F F pα1 αm qF pβ1 βn q.
¸
αi βj „ „rmsrns
¸
¸
¸
L K
¸
n
¸
P
|P| |Q| |P||Q|
βj j Q
In the fifth step the lemmata were used to evaluate the inner sum.
120
Chapter 7. Deviations and Cross-Evects
Let n be a natural number, and let P, Q „ rns. Then
Lemma 3.
¸
„
„
Y r s
p1q|I |
|J | 0,
P I
Q J
I J n
unless P
Q.
Proof. Compute
¸
„
„
Y r s
p1q|I |
¸
|J | P I
Q J
I J n
¸
„ „rns Q„J „rns
rnszI „J
p1q|I |
|J |
P I
¸
„ „rns
¸
p1q|I |
P I
Q
YprnszI q„J „rns
p1q|J | ,
and note that the inner sum vanishes unless Q Y prnszI q rns, or, equivalently,
I „ Q. Hence
¸
¸
p1q|I | |J | p1q|I | n 0,
„
„
Y r s
„„
P I
Q J
I J n
when P
Q.
P I Q
Let n be a natural number, and let P
Lemma 4.
¸
„
Y r s
p1q|I |
„ rns. Then
|J | p1qn|P| .
P I ,J
I J n
Proof. By the preceding proof,
¸
„
Y r s
p1q|I |
|J | P I ,J
I J n
¸
p1q|I | n p1qn|P| .
I P
Theorem 2.
pF b Gqpα1 αn q ¸
Y rns
I J
Proof. Compute:
¸
Y rns
I J
F
♦ αi
P
i I
bG
♦ αj
P
j J
F
♦ αi
P
i I
bG
♦ αj .
P
j J
§1. The Deviations
¸
„
Y rns
¸
P ,Q
„rns
P
„rns
¸
P
p1q|P|
| Q| F
¸
P
„rns
b p1q|J ||Q| G i P
¸
P
b G
αi
i P
p1qn|P| pF b Gq
„
Q J
¸
P
¸
P
αj αi
b G
i P
¸
P
αi
¸
P
αj j Q
j Q
p1q|P| |P| p1qn|P| F
¸
αi
¸
p1q|I ||P| F
P I
I J
¸
121
¸
„
„
Y r s
p1q|I |
|J |
P I
Q J
I J n
¸
P
αj j P
pF b Gqpα1 αn q.
i P
In the fourth step, the lemmata were used.
Let X be a set. We shall write
M CX
to indicate that M „ 2X (that is, M is a family of subsets of X), such that M
for any Y „ X. Equivalently,
¤
M
¤
P
Z X.
Z M
Let n be a natural number, and let P
Lemma 5.
¸
„ rs
„ rns. Then
p1q|J | 0,
P JC n
unless P is of the form 2Q , for some Q „ rns.
”
Proof. If P
€2
P
”
, choose A P 2
iv it satisfies
P
zP. Because the set J satisfies
P „ J C rns
P
„ J Y tAu C rns,
sets J with and without A will cancel each other out in the sum.
Lemma 6.
Let n be a natural number, and let Q „ rns. Then
¸
„ rs
2Q J C n
p1q|J | p1qn|Q|2|
|
Q
.
† 2Y
122
Chapter 7. Deviations and Cross-Evects
Proof. The formula is clearly valid when n |Q|. Now assume n
consider for the moment the sets J satisfying
J zp2rn1s Y ttnuuq ∅.
PJ
(There is thus a set Y
satisfies
¡ |Q|, and
P Y , but Y tnu.)
with n
„ J C rns
2Q
iv it satisfies
2Q
Because such a set J
„ J Y tnu C rns,
sets J with and without tnu will cancel each other out.
There remain the sets J with
J
„ 2rn1s Y ttnuu.
Since J C rns, necessarily tnu P J, and we may write J K Yttnuu for K
We then have
¸
¸
p1q|J | p1q|K| 1 ,
„ rs
„ r s
2Q J C n
2Q K C n 1
and the formula follows by induction.
Theorem 3.
¸
pF Gqpα1 αn q rs
F
♦ αi
♦G
P
P
I J
JC n
.
i I
Proof. Compute:
¸
rs
F
JC n
♦G
P
I J
¸ ¸
rs „
♦ αi
P
i I
p1q|J ||P| F
JC n P J
¸
P 2rns
„
¸
Q
„rns
¸
Q
„rns
p1q|P| F
¸
P
P
G
|
Q
F
¸
P
I 2Q
♦ αi
P
i I
I P
p1q2|
G
I P
¸
P
„ rs
i I
♦ αi
G
p1qn|Q| F G pF Gqpα1 αn q.
P
i I
¸
P
i Q
In the fourth step, the lemmata were used.
p1q|J |
P JC n
¸
♦ αi
αi p1qn|Q|2|
|
Q
„ 2rn1s .
§2. The Cross-Evects
123
§2. The Cross-Effects
An arbitrary module functor may be analysed in terms of its cross-effects. We
here indicate how this is done. Again, no assumptions are placed on either the
functors or the base ring.
Example 1.
Consider the second symmetric functor S2 , and let X and Y be
modules. We see that
S2 pX q xx1 x2 | xi P X y
contains the monomials obtained from X, and
S2 pY q xy1 y2 | yi
P Yy
the monomials from Y only. If S2 were additive, we would have
S 2 pX ` Y q S 2 pX q ` S 2 pY q.
But not so; mixed terms of the form xy will appear in S2 pX ` Y q, terms that
were not present in either S2 pX q or S2 pY q. This is the cross-effect pS2 q: pX | Y q. 4
The cross-evects may be defined as either of four equally canonical modules. Given a direct sum M M1 ` ` Mn , let
πj : M
ÑM
be projection on the jth summand,
ρj : M
Ñ M {Mj
retraction from the jth summand, and
τj : M {Mj
ÑM
insertion of 0 into the jth summand.
For a module functor F, the following four modules are naturally iso-
Theorem 4.
morphic:
A Im F pπ1 πn q : F pM q Ñ F pM q
B Ker pF pρ1 q, . . . , F pρn qq : F pM q Ñ
à
à
F pM {Mj q
Coker F pτ1 q F pτn q : F pM {Mj q Ñ F pM q
D Coim F pπ1 πn q : F pM q Ñ F pM q .
C
Proof. We only show that modules A B, and leave the remaining cases to the
reader.
Suppose
z P KerpF pρ1 q, . . . , F pρn qq.
124
Chapter 7. Deviations and Cross-Evects
Note that if j
i, then πi τj ρj πi . Consequently, if j R I, then
F
¸
P
πi
pz q F
¸
P
i I
πi
F pτj qF pρj qpzq 0.
i I
It follows that
¸
F pπ1 πn qpzq I
p1q |I | F
n
„rns
¸
P
πi
pz q
i I
F pπ1 πn q pzq F p1qpzq z.
Conversely, assume
z F pπ1 πn qpyq P Im F pπ1 πn q.
Then, since ρj πj
0, we get
¸
F pρj qpzq F pρj qF pπ1 πn qpyq ¸
I
„rns
p1qn|I | F I
¸
P zt u
„rns
p1qn|I | F
ρj
¸
P
πi
pyq
i I
ρj πi pyq 0,
i I j
because sets I with and without j will give cancelling terms.
Definition 1.
The pn 1qst cross-effect of F is the multi-functor
F : pM1
| | Mn q
of n arguments, defined as any of the four modules above.
For our purposes, it shall be most convenient to view the cross-evect as
F : pM1
| | Mn q Im F pπ1 πn q.
It is also the definition which generalises most readily to yield the multi-crosseffects of a strict analytic functor.
Let us now describe, for given homomorphisms αi : Mi Ñ Mi1 , how to form
the map
F : pα1
| | αn q : F : pM1 | | Mn q Ñ F :
M11
| | Mn1
It is easy to verify the commutativity of the following diagram:
À
Fp
p
F ♦ πi
pÀ α q / F pÀ M 1 q
i
i
q
À
Fp
Mi q
F
Mi q
F
pÀ α q
i
p q
/ F p M 1q
i
À
F ♦ π1i
.
§2. The Cross-Evects
125
Therefore, there will be an induced homomorphism
Im F p♦ πi q Ñ Im F ♦ π1i ,
which we define to be F : pα1 | | αn q. It will now be readily checked that, in
fact,
F : pα1 | | αn q F pα1 αn q.
When presented with a (finite) family pMi qiPI of modules, we shall write
F : pMi |I q
for the cross-evect of the modules Mi .
Theorem 5: The Cross-Effect Decomposition.
à
F pM1 ` ` Mk q I
„rks
F : pMi |iPI q.
Proof. See [6].
This decomposition is evidently functorial, and there is a corresponding
decomposition of natural transformations ζ : F Ñ G:
Theorem 6.
à ζM1 ``Mk
I
„rks
ζ:M |
P
i i I
: F : pMi |iPI q Ñ G: pMi |iPI q .
F is polynomial of degree n iff its nth cross-effect vanishes.
Proof. Suppose the nth cross-evect vanishes, and consider n
isms
α1 , . . . , αn 1 : M Ñ N.
Let Mj
M and Nj N, for j 1, . . . , n
πj :
à
Ni
1, let
Ñ
à
Ni
denote the jth projection, and define
σ:
à
Ni
1 homomorph-
Ñ N, py1 , . . . , yn 1 q ÞÑ
¸
yi .
The equality
F pα1 αn
1
q F pσq F pπ1 πn 1 q F ppα1 , . . . , αn 1 qq
is easily checked, for maps
F pN q Ð F
à
Ni
ÐF
à
Ni
Ð F pM q.
Since the middle component is zero by assumption, we conclude that
F pα1 αn
and so F is polynomial of degree n.
The other direction is trivial.
1
q 0,
126
Chapter 7. Deviations and Cross-Evects
Example 2.
For any functor F,
F : pq F p0q
is the “constant term” of F.
If F is reduced, that is, F p0q 0; then the zeroth cross-evect coincides with
the functor itself:
F : pX q F pX q.
In the general case,
F pX q F : pq ` F : pX q.
4
Example 3.
If F is polynomial of degree 1, then all cross-evects of order
higher than 0 vanish, and
F pX ` Y q F : pq ` F : pX q ` F : pY q.
4
Example 4.
1 vanish, and
For a functor of degree 2, all cross-evects of order higher than
F pX ` Y q F : pq ` F : pX q ` F : pY q ` F : pX | Y q.
To take a concrete example, we have for S2 :
pS2 q: pq 0
pS2 q: pX q xx1 x2 | xi P X y
pS2 q: pY q xy1 y2 | yi P Y y
pS2 q: pX | Y q xx1 y1 | xi P X, yi P Y y .
The equation
pS2 q: pX | Y | Zq 0
allows for the following interpretation in words: there are no monomials of
degree 2 involving elements from all three modules X, Y , and Z.
4
§3. The Multi-Deviations
The deviations, which exist for arbitrary maps and functors, find for strict
polynomial maps and functors their equivalent in the multi-deviations, introduced earlier. We shall derive formulæ for the composition of multi-deviations, the multi-deviations of a tensor product, and the multi-deviations of a
composition. Perhaps surprisingly, the proofs will here be substantially shorter, because no combinatorial tinkering will be necessary.
§3. The Multi-Deviations
127
Throughout this section and the next, F and G will denote strict analytic
functors over a commutative base ring. Let αi : M Ñ N be homomorphisms.
Recall that the maps
FαrAs : F pM q Ñ F pN q
are defined by the universal validity of the equation
F
¸
ai b αi
¸
i
aA b FαrAs .
A
We have mentioned the formula
xrAs yrBs
¸
µ: A
ÑB
pxyqrµs
for the product multiplication on the divided power algebras. Its functorial
generalisation is as follows.
Theorem 7.
FαrAs
Fβr s ¸
B
µ: A
ÑB
Fpαβqrµs .
Proof. Identify the coeHcient of aA bB in
¸
aA b FαrAs
¸
bB b FβrBs
F
B
A
¸
F
ai b αi
F
i
¸
¸
bj b βj j
ai bj b αi βj i, j
Theorem 8.
¸
µ: A
¸
pF b Gqαr s X
\
ÑB
FαrAs
pabqµ b Fpαβqr s .
µ
b Gαr s .
B
A B X
Proof. Identify the coeHcient of aX in
pF b Gq
¸
ai b αi
F
i
¸
ai b αi
i
¸
A
a
bG
A
b F αr s b
¸
ai b αi
i
¸
A
B
a
B
b Gαr s
B
.
128
Chapter 7. Deviations and Cross-Evects
P X, be homomorphisms.
Let G be a functor, and let αk , for k
partition of X, we define the symbol
rE s ä G
αrY s
Y PE
Gα
Thus, for example, if
is a partition of X
¹
If E is a
rdeg Y s .
P
Y #E
GαrY sE
E tt1, 1u, t2, 3u, t2, 3uu
t1, 1, 2, 2, 3, 3u, then
rEs Gr1s
αrt1,1us
Gα
Theorem 9.
r2 s
G rt2,3us
α
Gαr s Grα2sα .
1
¸
pF Gqαr s X
P
E Par X
2
2 3
FGrEs .
α
Proof. Identify the coeHcient of aX in
pF Gq
¸
ai b αi
F
i
F
G
¸
¸
aY
Y
ai b α i
i
¸ ¸
b Gαr
P
¹
P
X E Par X
Y
s
a
Y
b FÄ
Y E
P
Y E
G rY s .
α
§4. The Multi-Cross-Effects
The cross-evects of a strict analytic functor may be further dissected into so
called multi-cross-effects. Let us begin with an example to illustrate the concept.
Example 5.
Let X and Y be modules, and consider the third symmetric
functor S3 . We know from before that the cross-evect pS3 q: pX | Y q contains
the cubic monomials built by elements of both X and Y . This module splits
canonically into two components:
pS3 q: pX | Y q @
x1 x2 y1 | xi
D
P X, yj P Y `
@
x1 y 1 y 2 | xi
P X, yj P Y
D
.
These are the multi-cross-evects
pS3 q:tX ,X ,Y u pX | Y q
respectively.
and
pS3 q:tX ,Y ,Y u pX | Y q,
4
§4. The Multi-Cross-Evects
129
When given a direct sum
M
M1 ` ` Mn ,
πi : M Ñ M will as usual denote the ith projection. We recall that the crossevects of an arbitrary module functor F are given by
F : pM1
| | Mn q Im F pπ1 πn q.
Let A be a multi-set with #A „ rns. The multi-cross-effect
Definition 2.
of F of type A is the multi-functor
FA: pM1
| | Mn q Im Fπr s
A
of n arguments.
When pMi qiPI is a family of modules, we may write
FA: pMi |I q
for the multi-cross-evect of type A of the modules Mi .
Theorem 10: The Multi-Cross-Effect Decomposition.
F : pM1
| | Mn q à
#A
rns
FA: pM1
| | Mn q ,
FA: pM1
| | Mn q .
and consequently,
F pM1 ` ` Mn q à
#A
„rns
Proof. It immediately follows from the equation
F
¸
ai b π i
¸
i
that the identity map on
aA b FπrAs
A
F pM1 ` ` Mn q
decomposes as
1 F p1q F
¸
πi
i
¸
FπrAs .
A
Furthermore, the equation
¸
A, B
aA bB b FπrAs FπrBs
F
¸
i
ai b πi
F
¸
j
bj b πj 130
Chapter 7. Deviations and Cross-Evects
F
¸
ak bk b πk
C
C
k
shows that
FπrAs FπrBs
#
¸
pabqC b Fπr s
when A B,
when A B.
0
FπrAs
Consequently, the maps FπrAs provide a direct sum decomposition.
This decomposition is evidently functorial, and there is a corresponding
decomposition of natural transformations ζ : F Ñ G:
ζM1 ``Mn
à
#A
„rns
pζ:A qM ||M .
1
n
Note in particular the following special case of the theorem:
F pBn q à
#A
FA: pB|n q,
„rns
which is the one that will be most frequently used.
Theorem 11.
F is homogeneous of degree n iff its multi-cross-effects of type A vanish
whenever |A| n.
Proof. If F is homogeneous of degree n, then plainly
FA: pM1
| | Mn q Im Fπr s 0
A
only if |A| n.
Conversely, suppose that the multi-cross-evects of F of type A vanish when
|A| n. Because there is a direct sum decomposition
1F pM q
F p1M q ¸
m
and we made the assumption that
Im F1rms
M
0
when m n, we deduce that, in fact,
F1rms
M
when m n, and hence
Let α : M
1F pM q
0
F1r s .
n
M
Ñ N be a homomorphism. We get
F pa b αq F p1 b αqF pa b 1M q
F1rms ,
M
§4. The Multi-Cross-Evects
131
p1 b F pαqq
¸
m
n
am b F1rms
M
p1 b F pαqqpa b 1F pM q q an b F pαq.
Of course, F is strict polynomial of degree n iv the multi-cross-evects of
type A vanish whenever |A| ¡ n.
Example 6.
When F is homogeneous of degree 0, then
: pq F : pq F p0q,
Ftu
with all other multi-cross-evects vanishing.
Example 7.
4
When F is homogeneous of degree 1, then
Ft:X u pX q F : pX q F pX q
are the only non-vanishing multi-cross-evects.
4
Example 8.
Even in the case of homogeneity degree 2, the multi-crossevects coincide with the cross-evects, and thus provide no further decomposition. So, for example:
Ft:X ,X u pX | Y q F : pX q
Ft:Y ,Y u pX | Y q F : pY q
Ft:X ,Y u pX | Y q F : pX | Y q.
4
Example 9.
The utility of the multi-cross-evects becomes apparent once
we reach homogeneity degree 3. In this case, we have
F : pX q Ft:X ,X ,X u pX | Y | Zq
F : pX | Y q Ft:X ,X ,Y u pX | Y | Zq ` Ft:X ,Y ,Y u pX | Y | Zq
F : pX | Y | Zq Ft:X ,Y ,Zu pX | Y | Zq.
We see that F : pX
example.
| Y q all of a sudden decomposes, as in the introductory
4
Example 10.
As our final, very general, example, we choose the complete
symmetric functor S. Abbreviating
@
P X, yj P Y
we obtain the following decomposition of SpX ` Y q:
X pY q
x1 . . . xp y1 . . . yq | xi
D
,
132
Chapter 7. Deviations and Cross-Evects
S: pq
S : pX q
S: pY q
S : pX | Y q
$
'
'
'
'
&
'
'
'
'
%
$
'
'
'
'
&
'
'
'
'
%
$
'
'
'
'
'
'
'
'
'
'
'
'
&
'
'
'
'
'
'
'
'
'
'
'
'
%
: pX | Y q
Stu
:
StX u pX | Y q
St:X ,X u pX | Y q
St:X ,X ,X u pX | Y q
1
X
X2
X3
St:Y u pX | Y q
St:Y ,Y u pX | Y q
St:Y ,Y ,Y u pX | Y q
Y
Y2
Y3
XY
X 2Y
XY 2
X 3Y
X 2Y 2
XY 3
St:X ,Y u pX
St:X ,X ,Y u pX
St:X ,Y ,Y u pX
St:X ,X ,X ,Y u pX
St:X ,X ,Y ,Y u pX
St:X ,Y ,Y ,Y u pX
| Yq
| Yq
| Yq
| Yq
| Yq
| Yq
..
.
..
.
..
.
4
Chapter 8
PROJECTIVE
GENERATORS
The first person he met was Rabbit.
“Hallo, Rabbit,” he said, “is that you?”
“Let’s pretend it isn’t,” said Rabbit, “and see what happens.”
”I’ve got a message for you.”
“I’ll give it to him.”
“We’re all going on an Expotition with Christopher Robin!”
“What is it when we’re on it?”
“A sort of boat, I think,” said Pooh.
“Oh! that sort.”
“And we’re going to discover a Pole or something. Or was it a Mole? Anyhow
we’re going to discover it.”
“We are, are we?” said Rabbit.
“Yes. And we’ve got to bring Pro-things to eat with us. In case we want to eat
them. [. . . ]”
Alan Alexander Milne, Winnie-the-Pooh
All functors are equal, but some are more equal than others. Indeed, some
functors turn out to possess the pleasant property of being a projective generator of the category they live in. Expounding upon these amiable functors is
the purpose of the present chapter.
If the theory of polynomial functors be likened to a palace, then this chapter, along with the next, could not unjustly be called the oriental garden encircling it. Japanese influences are here highly perceptible. The course of the
current chapter will have us indulge in a veritable orgy of lemmata, which
all bear the unmistakable mark of Yoneda, while the theorems of the next
are elaborate instances of the classical Morita Equivalence. Of course, the
counterexamples, if there were any, would probably be due to Nagata.
§1. The Fundamental Module Functor
Let us begin by exhibiting a projective generator for the category of all module
functors. As usual, B is then permitted to be an arbitrary, unital ring.
Theorem 1.
Let K be a fixed module. The functor BrHompK, qs, given by
M
ÞÑ BrHompK, M qs
133
134
Chapter 8. Projective Generators
χ: M
M qs Ñ BrHompK, N qs
Ñ N ÞÑ rχ s : BrHomrpαK,s ÞÑ
rχ αs
,
is a module functor.
Define
B8
8
à
n 1
B
8
¤
Bn
n 1
lim
Bn ,
Ý
Ñ
nÑ8
the module of all finite vectors over B. The functor
BrHompB8 , qs
will be called the fundamental module functor. It is a projective generator for
the category of module functors, as will be demonstrated in the next chapter.
§2. The Classical Yoneda Correspondence
Recalling the classical Yoneda correspondence, we strongly emphasise that
natural transformations are always assumed linear, whereas functors are not.
Theorem 2: The Classical Yoneda Lemma.
functor. The map
ΥK ,F : Nat BrHompK, qs, F
η ÞÑ ηK pr1K sq
Let K be a fixed module, and F a
Ñ F pK q
is an isomorphism of modules.
The isomorphism is natural, in the sense that the following two diagrams commute:
K
β
NatpBrHompK, qs, F q
rβ s
L
NatpBrHompL, qs, F q
F
NatpBrHompK, qs, F q
ξ
G
ΥK,F
ΥL,F
ΥK,F
ξ
NatpBrHompK, qs, Gq
/ F pK q
pq
F β
/ F pLq
/ F pK q
ΥK,G
ξK
/ GpK q
Proof. This is, in fact, the original Yoneda Lemma in disguise. Note that arbitrary natural transformations
HompK, q Ñ F
correspond to linear transformations
BrHompK, qs Ñ F.
§3. The Fundamental Numerical Functor
135
§3. The Fundamental Numerical Functor
We now turn to numerical functors, and, as is customary, assume a numerical
base ring.
Let K be a fixed module. The functor BrHompK, qsn , given by
Theorem 3.
ÞÑ BrHompK, M qsn
r
χ s : BrHompK, M qsn Ñ BrHompK, N qsn
χ : M Ñ N ÞÑ
,
rαs ÞÑ rχ αs
M
is numerical of degree n.
Proof. Since BrHompK, qsn is the composition of Brsn with the Hom-functor, it suHces to prove Brsn is of degree n. Let χj : M Ñ N be homomorphisms, and let x P M; then
rχ1 χn 1 sprxsq rχ1 pxq χn 1 pxqs 0.
Moreover, if a P B and χ : M Ñ N, then
raχsprxsq raχpxqs ņ
k 0
a
k
♦ χpxq
k
k 0
We infer that Brsn is numerical of degree n.
The case K
ņ
a
k
♦χ
k
prxsq.
Bn is especially important. The functor
BrHompBn , qsn
will be called the fundamental numerical functor of degree n. It will presently
be seen to be a projective generator.
Example 1.
We give an example of a functor which is polynomial, but not
numerical, of degree 1. Let the base ring be R, and define for real vector spaces
F : R Mod Ñ R Mod
V ÞÑ RrV s{ xrx ys rxs rysy .
We thus impose additivity, but not linearity. Since F is additive, its first crossevect will vanish. But F is not numerical of degree 1, for
?
and r 2s ?2r1s in
r?2s
p?2 : R Ñ Rq : r1s ÞÑ ?
?F2F
p1 : R Ñ Rq : r1s ÞÑ 2r1s,
F pRq RrRs{ hrx
In fact, F is not numerical of any degree.
ys rxs rysi .
4
136
Chapter 8. Projective Generators
§4. The Numerical Yoneda Correspondence
Let K be a fixed module, and F a
Theorem 4: The Numerical Yoneda Lemma.
numerical functor of degree n. The map
ΥK ,F : Nat BrHompK, qsn , F
η ÞÑ ηK pr1K sq
Ñ F pK q
is an isomorphism of modules.
The isomorphism is natural, in the sense that the following two diagrams commute:
NatpBrHompK, qsn , F q
K
ΥK,F
/ F pK q
rβ s
β
L
NatpBrHompL, qsn , F q
F
NatpBrHompK, qsn , F q
ΥL,F
/ F pLq
ΥK,F
/ F pK q
ξ
ξ
NatpBrHompK, qsn , Gq
G
pq
F β
ξK
/ GpK q
ΥK,G
Proof. The proof is the usual one. Consider the following commutative diagram:
K
α
M
ηK
BrHompK, K qsn
rα s
/ F pK q
pq
F α
BrHompK, M qsn
ηM
/ F pM q
r1K s
/ ηK pr1K sq
/ ηM prαsq F pαqpηK pr1K sqq
rαs
Upon inspection, we find that ΥK ,F has the inverse
y ÞÑ
ηM : BrHompK, M qsn Ñ F pM q
.
rαs ÞÑ F pαqpyq
When defining this inverse, the numericality of F is used in an essential way
to ensure that the map
HompK, M q Ñ HompF pK q, F pM qq
factor through BrHompK, M qsn .
The naturality of Υ is obvious.
§5. The Deviated Power Functors
137
In particular, putting F BrHompK, qsn , we obtain a module isomorphism
NatpBrHompK, qsn q BrHompK, K qsn BrEnd K sn ,
given by the map
with inverse
Υ : η ÞÑ ηK pr1K sq
rσ s : BrHompK, qsn Ñ BrHompK, qsn .
Υ1 : rσs ÞÑ
rαs ÞÑ rα σs.
We note that
Υ1 prσsrτsq Υ1 prσ
rσ
τq s
τsq rpσ
τ s rσ srτ s Υ1 prσsqΥ1 prτsq,
so that the Yoneda correspondence is actually a ring isomorphism under the
sum multiplication.
Now, this is probably very interesting and all, but for our purposes it will
be of no major consequence, and we mentioned the above fact just in passing.
The really interesting question is what happens to the product multiplication
on BrEnd K sn :
Υ1 prσs rτsq Υ1 prστsq rpστq s
rτ s rσ s Υ1 prτsq Υ1 prσsq.
The product multiplication is reversed by Υ.
Theorem 5.
The Yoneda correspondence provides an isomorphism of rings
Nat BrHompK, qsn
BrEnd K sn ,
where the former is equipped with composition, and the latter with the product multiplication.
§5. The Deviated Power Functors
The fundamental numerical functor
BrHompBn , qsn
is not atomic, but built from simpler components. A similar decomposition
is available for the functor
BrHompB8 , qs,
and, in order to subsume this case, we shall, in what follows, allow for the
possibility n 8. The symbol
Num8
138
Chapter 8. Projective Generators
will then denote arbitrary module functors, and we agree that
Laby8
Laby.
In the case n 8, numericality of the base ring shall of course not be needed.
We draw the reader’s attention to a little peculiarity. When including the
case n 8, we allow for a non-commutative base ring B. However, since
we limit ourselves to the study of finitely generated and free modules, all our
modules are automatically bimodules. Homomorphisms are still one-sided, and
we settle on the consideration of right module homomorphisms only, as this turns
out to give the most natural theory.
When pMi qiPI is a finite family of modules, we define
♦ Mi
P
i I
n
B
F
♦ ui
P
i I
P Mi „ B
ui
à
P
Mi
i I
.
n
As a degenerate case, we have
xrsy xr0sy B.
♦ Mi
P
i ∅
n
Let X be a finite set. Define the corresponding deviated power functor by
∆
X
pM q ♦ M
P
x X
n
„B
à
P
x X
M
.
n
The deviated power category Dev is the category of functors
Definition 1.
∆X , where X P Labyn .
The definition we have given of deviated powers, in terms of deviations, is
algebraic, but awkward. It turns out that the most natural, and fruitful, way
of viewing them is as mazes.
Let M be a module, and X and Y be finite sets. We will denote by
M Labyn
pX, Y q
the module of mazes X Ñ Y , where the passages have been labelled with
elements of M (rather than B). We can, and will, still impose the labyrinth
axioms — two if n 8, and four if n 8. It should be noted that there is no
way of composing such mazes (as there is no multiplication in M), so we do
not obtain a category.
Evidently, the assignment
M
ÞÑ M Labyn pX, Y q
Ñ N, we shall denote by
α Labyn pX, Y q : M Labyn pX, Y q Ñ N Labyn pX, Y q
is functorial. When α : M
§5. The Deviated Power Functors
139
the induced homomorphism (acting on labels of passages).
Putting Y , the canonical one-element set, will provide us with the
sought description of the deviated power functors. We are forced to accept
the degenerate case X ∅. An element of M Labyn p∅, q will be an (unlabelled)
“dead end”, which we otherwise took great care to forbid, but it is necessary
to allow this pathology in order to handle the functor ∆∅ .
Let two direct sums M X and M Y be given. Recall that, for p P X and q P Y ,
there is a canonical transportation map
σqp : M X
Theorem 6.
Ñ MY ,
σqp ppux qxPX qy
#
if y q
if y q.
0
up
Let X be a finite set, and M a module. The map
ΞX ,M : ∆X pM q Ñ M Labyn pX, q
¤
ÞÑ
♦ ux
P
x X
"
ux
x
P
/
*
x X
is an isomorphism of modules.
The isomorphism is natural, in the sense that the following two diagrams commute:
∆X pM q
X
r♦r
P
Y
p: x
ÑysPP pσyx s
∆Y pM q
∆X pM q
M
r αs
α
N
Proof. Define a map
B
à
P
Ñ
M
x X
rpux qxPX s ÞÑ
ΞX,M
∆X pN q
/ M Labyn pX, q
ΞY ,M
/ M Labyn pX, q
ΞX,N
à
„
¤
P
z Z
"
z
pP q
/ M Labyn pY , q
pq
Labyn α
/ N Labyn pX, q
M Labyn
Z X
n
ΞX,M
uz
/
pX, q
*
,
„
Z X
and note that ΞX ,M is the restriction of this to ∆X pM q. Hence Ξ is well-defined,
and it has an obvious inverse.
The second diagram above evidently commutes. That the first one does
is a consequence of the Deviation Formula and the definition of the maze
product.
140
Chapter 8. Projective Generators
Example 2.
The maze
x vv 1
vv
{v{vvvy cHH P M Labyn pt1, 2u, q
H
z
2
corresponds to the deviation
rpx, 0q py, 0q p0, zqs P ∆2 pM q.
4
We see that deviations of module elements correspond to mazes. That also
deviations of homomorphisms can be naturally interpreted as mazes should
come as no surprise.
Theorem 7.
Given homomorphisms αk : M
S
¤
"
αk
/
*
Ñ N, let
P HompM ,N q Labyn p, q.
The following diagram commutes:
∆X pM q
r♦ α s
k
ΞX,M
∆X pN q
/ M Labyn pX, q
ΞX,N
S
/ N Labyn pX, q
Proof. The formula
¸
rα1 αm s pru1 un sq K
♦ αi puj q
„rmsrns pi,jqPK
can be obtained as a special case of the Deviation Formula (it is, in any case,
proved in exactly the same way). The commutativity of the above diagram is
then an immediate consequence.
Let P : X
Ñ Y be a maze. We denote by
∆P : ∆Y Ñ ∆X
the corresponding natural transformation
P : Labyn pY , q Ñ Labyn pX, q.
With the current description of deviated powers, the verification of the
following theorem becomes a mere triviality.
§6. The Labyrinthine Yoneda Correspondence
Theorem 8.
141
The functor
Ξ : Labyn
Ñ Devn ,
mapping
Þ ∆X
Ñ
rP : X Ñ Y s ÞÑ ∆P : ∆Y Ñ ∆X ,
X
is an anti-isomorphism of categories.
Proof. That every transformation
M Labyn
pY , q Ñ M Labyn pX, q,
which is natural in M, is of the form P , and uniquely so, is not diHcult to see.
Recall that we may choose M to be a free algebra.
Example 3.
We have thus made three important identifications. The set
t1, 2u P Labyn corresponds under the category anti-isomorphism to the functor ∆2 , which, as we know, is naturally isomorphic to the functor
Labyn pt1, 2u, q.
In what follows, such identifications will be made without comment.
4
The importance of the deviated power functors stems from the fact that
they provide a splitting of the fundamental (numerical) functor into atomic
components (and here, for once, the original definition is actually useful). To
subsume the case n 8, we put r8s Z .
Theorem 9.
BrHompBn , qsn
à
X
„rns
∆X .
Proof.
BrHompB , M qsn
n
B
à
Pr s
k n
M
n
à
♦ M
X „rns P
x X
n
à
X
„rns
§6. The Labyrinthine Yoneda Correspondence
As usual, σ denotes transportation maps.
∆X pM q.
142
Chapter 8. Projective Generators
Let X be a set, and F a func-
Theorem 10: The Labyrinthine Yoneda Lemma.
tor. The map
ΥX ,F : Natp∆X , F q Ñ F : pB|X q
ζ ÞÑ ζBX
"
¤
1x
x
P
/
*
x X
is an isomorphism of modules.
The isomorphism is natural, in the sense that the following two diagrams commute:
Natp∆X , F q
X
ΥX,F
/ F : pB|X q
p
p∆P q
P
Natp∆Y , F q
Y
F ♦rp : xÑysPP pσyx
/ F : pB|Y q
ΥY ,F
Natp∆X , F q
F
ΥX,F
Natp∆X , Gq
G
/ F : pB|X q
ξ:B|
ξ
ξ
q
X
ΥX,G
/ G: pB|X q
Proof. A natural transformation
ζ : ∆X
ÑF
corresponds to a natural transformation
ηM : BrHompBX , M qs Ñ F pM q,
taking
♦ rαy : By
P
y Y
Ñ M s ÞÑ
$
&0 ”
%ζM
yPY
o
p q
αy 1y
€ X,
if Y X.
if Y
y
The original Yoneda map takes η to
ηBX pr1BX sq ¸
„
ηBX
Y X
ζB
X
¤
P
x X
♦ 1By
P
y Y
p q
x
o
1Bx 1x
ηB
♦ 1Bx
X
P
x X
ζB
X
¤
P
x X
o
1x
x
.
§6. The Labyrinthine Yoneda Correspondence
143
Thus, ΥX ,F is the X-component of the original Yoneda map, and it is clear that
ΥX ,F pζq ηBX
P F : pB|X q.
♦ 1Bx
P
x X
That the second diagram above commutes is clear, so there remains to
investigate the first. From the original Yoneda Lemma, there is a commutative
diagram:
NatpBrHompBX , qs, F q
ÑysPP pσyx q
p♦r
p: x
ΥBX ,F
/ F pBX q
NatpBrHompBY , qs, F q
p
F ♦rp : xÑysPP pσyx
/ F pBY q
ΥBY ,F
q
We shall show that the following diagram commutes, from which the claim
follows:
/ NatpBrHompBX , qs, F q
Natp∆X , F q
p♦r
p∆P q
p: x
ÑysPP pσyx q
/ NatpBrHompBY , qs, F q
Natp∆Y , F q
Consider a natural transformation
Ñ F,
ζ : ∆X
corresponding to an
η : BrHompBX , qs Ñ F
as above. The homomorphism
♦
rp : xÑysPP
pσyx
maps η to the natural transformation
ηM
♦
rp : xÑysPP
pσyx
which takes
♦ rαy : By
P
y Y
Ñ M s ÞÑ ηM
ηM
: BrHompBY , M qs Ñ F pM q,
♦ αy P
y Y
♦
♦
rp : xÑysPP
rp : xÑysPP
pσyx
αy pσyx
144
Chapter 8. Projective Generators
ζM ζM ¤
rp : xÑysPP
¤
rp : xÑysPP
o
p
p qq
αy pσyx 1x
o
p q
αy 1 y p
x
x
,
where the last step is because
αy ppσyx p1x qq αy pp 1y q αy p1y pq αy p1y q p.
On the other hand,
ÑF
ζ∆P : ∆Y
corresponds to a natural transformation
BrHompBY , M qs Ñ F pM q,
that takes
♦ rαy : By
P
y Y
Ñ M s ÞÑ ζM ¤
P
y Y
o
p q
αy 1y
y
¤
rp : xÑysPP
yo
p
x
.
The commutativity of the diagram is demonstrated.
§7. The Fundamental Homogeneous Functor
We recall that, given ai P A (where A is some algebra) and homomorphisms
αi P HompM, N q, the equation
Γ
n
¸
ai b αi
i
¸
|X |n
aX
b pΓn qαr s
X
defines the multi-deviations
pΓn qαr s : Γn pM q Ñ Γn pN q.
X
As a matter of notational convenience, let us agree to write
αrX s
pΓn qαr s .
X
The symbol αrX s may thus be interpreted either as an element of
Γn HompM, N q
or as a map
(and usually as both).
Γn pM q Ñ Γn pN q
§8. The Homogeneous Yoneda Correspondence
Theorem 11.
145
Let K be a fixed module. The functor Γn HompK, q, given by
ÞÑ Γn HompK, M q
pχ qrns : Γn HompK, M q Ñ Γn HompK, N q ,
χ : M Ñ N ÞÑ
αrns ÞÑ pχ αqrns
M
is homogeneous of degree n.
It will prove convenient to have available a formula for the strict polynomial structure. Thus, for any algebra A:
°
i ai b χ i
P A b HompM, N q
Again, the case K
ÞÑ
rns
°
°
|X |n aX b pχ qrX s .
i ai b pχi q
n
P A b HompΓ HompK, M q, Γn HompK, N qq
Bn is especially important. The functor
Γn HompBn , q
will be called the fundamental homogeneous functor of degree n. It will
presently be seen to be a projective generator.
§8. The Homogeneous Yoneda Correspondence
Let K be a fixed module, and
Theorem 12: The Homogeneous Yoneda Lemma.
F a homogeneous functor of degree n. The map
ΥK ,F : NatpΓn HompK, q, F q Ñ F pK q
r ns q
η ÞÑ ηK p1K
is an isomorphism of modules.
The isomorphism is natural, in the sense that the following two diagrams commute:
K
β
NatpΓn HompK, q, F q
ppβ qrns q
L
NatpΓn HompL, q, F q
F
NatpΓn HompK, q, F q
ξ
G
ΥK,F
pq
F β
ΥL,F
ΥK,F
ξ
NatpΓn HompK, q, Gq
/ F pK q
/ F pLq
/ F pK q
ΥK,G
ξK
/ GpK q
146
Chapter 8. Projective Generators
Proof. Let y P F pK q, and consider the strict polynomial map
Ξ : HompK, M q
F
/ HompF pK q, F pM qq
/ F pM q,
where the second (linear) map is evaluation at y. Since this map is homogeneous of degree n, it gives rise to a linear map
αrns
Γn HompK, M q Ñ F pM q,
ÞÑ F pαqpyq.
We may therefore define
Ξ : F pK q Ñ NatpΓn HompK, q, F q
y ÞÑ
ζM : Γn HompK, M q Ñ F pM q
.
αrns ÞÑ F pαqpyq
It should be evident that ζ is indeed natural.
Let us now show that the above formula gives the inverse of Υ. On the one
hand, it is clear that
r ns q F p 1
ΥΞpyq ΞpyqK p1K
On the other hand, let
K
qpyq y.
η : Γn HompK, q Ñ F
be given. There is a commutative diagram:
K
α
M
Γn HompK, K q
ηK
pα qrns
/ F pK q
pq
rns
/ η p1rns q
K K
1K
F α
Γn HompK, M q
ηM
/ F pM q
αrns
/η
rn s
rn s
M pα q F pαqpηK p1K qq
We deduce that the natural transformation ΞΥpηq maps
α rn s
ÞÑ F pαqpΥpηqq F pαqpηK p1rKns qq ηM pαrns q,
and hence ΞΥpηq η.
The naturality of Υ is obvious.
Γn HompK, q, we obtain a module isomorphism
NatpΓn HompK, qq Γn HompK, K q Γn pEnd K q,
In particular, putting F
given by the map
with inverse
Υ1 : σrns
rn s
Υ : η ÞÑ ηK p1K q,
rn s n
pK, M q Ñ Γn HompK, M q
ÞÑ pσ q : Γ Hom
r
n
s
α ÞÑ pα σqrns
.
§9. The Divided Power Functors
147
As in the numerical case, we observe that
Υ1 pσrns τrns q Υ1 ppστqrns q ppστq q
rn s
pτ σ qrns pτ qrns pσ qrns Υ1 pτrns q Υ1 pσrns q,
and we have deduced the following theorem.
Theorem 13.
The Yoneda correspondence provides an isomorphism of rings
NatpΓn HompK, qq
Γn pEnd K q,
where the former is equipped with composition, and the latter with the product multiplication.
§9. The Divided Power Functors
When pMa qaPA is a (finite) multi-set of modules, we define
ä
P
Ma
a A
â
P
a #A
Γdeg a pMa q,
which might be called a confluent product of modules. As a degenerate case,
we have
ä
Ma B.
P
a ∅
Let A be a multi-set. Define the corresponding divided power functor by
ΓA pM q ä
P
a A
M
â
P
a #A
Γdeg a pM q.
Definition 2.
The nth divided power category Divn is the full subcategory
of Homn consisting of the functors ΓA , where A P MSetn .
In order to study the deviated powers, we took to establishing an isomorphism between ∆X pM q and a module of formal mazes, labelled with elements from M. Analogously, it will prove convenient to view elements of
ΓX pM q as formal multations. We recall that, at the very beginning, we saw
an example of the reverse procedure, that of viewing multations as divided
powers. There is indeed a very intimate connection between multations and
divided powers.
Let M be a module, let ξ : A Ñ K be a multation, and consider module
elements xk P M, for k P K. Recall that we have defined the symbol
xbrξs
â
P
ä
p qP
a # A a, k ξ
xk
P ΓA pM q.
If ξ is a sum of multations, we choose to interpret this quantity as the corresponding sum of divided powers. (This is why we prefer the symbol to the
148
Chapter 8. Projective Generators
left, as this could conceivably be linear in ξ, whereas it looks more doubtful
that the right one is.)
Consider now a multation µ : A Ñ B, and free variables xba . Let M be a
module. As usual, σba will denote transportation maps M #B Ñ M #A , but this
time acting on the right. Recall that the equation
rns
¸
xba
P
P
a #A
b #B
b σba ¸
µ
xµ b σrµs
defines homomorphisms
σrµs : Γn pM #B q Ñ Γn pM #A q.
(Here µ will range over multations X Ñ Y , where #X „ #A, #Y „ #B, and
|X | |Y | n.) We note that
ä
σrµs : ppxb qbP#B qrns ÞÑ
pxb qσba ,
pa,bqPµ
and hence σrµs may be viewed as a map
Γµ pM q : ΓB pM q Ñ ΓA pM q,
taking
xbrBs
ÞÑ xbrµs .
It is clear that Γµ is a natural transformation.
Example 4.
The multation
µ
1
1
1 2 2
2 2 2
provides a natural transformation
Γµ : Γt1,2,2,2u
Ñ Γt1,1,2,2u
r3s
r2s
x1 b x2 ÞÑ x1 x2 b x2 .
4
Theorem 14.
The functor
Ξ : MSetn
Ñ Divn ,
mapping
A ÞÑ ΓA
rµ : A Ñ Bs ÞÑ
is an anti-isomorphism of categories.
µ
Γ : ΓB
xbrξs
Ñ ΓA
ÞÑ xbrξµs
§9. The Divided Power Functors
149
Proof. Since Γµ is found as the coeHcient of xµ in
rns
¸
xba
P
P
a #A
b #B
b σba ,
while µ itself is the coeHcient of xµ in
¸
xba
P
P
b
a #A
b #B
rn s
a ,
b it is evident that the correspondence is functorial. Also, it should be clear that
any natural transformation ΓB Ñ ΓA can be expressed as a sum of maps Γµ ,
for, as usual, M may be chosen to be a free algebra.
Finally, let us verify the given formula for Γµ . Let ξ : B Ñ C be a multation;
then we already know that
Γξ : ΓC Ñ ΓB
maps
xbrC s
ÞÑ xbrξs .
Hence, by functoriality,
Γµ pxbrξs q Γµ pΓξ pxbrC s qq Γξµ pxbrC s q xbrξµs .
The theorem should be compared with the category anti-isomorphism
Setn
tT A | A P Setn u,
a result that appears to hark back to Weyl. Here, to complete the analogy,
Setn denotes the category of n-element sets and their bijections.
Dr. Salomonsson presents an alternative proof in [20]; more conceptual,
but with the disadvantage of obscuring the underlying combinatorics.
Example 5.
In the previous example, we studied the multation
µ
1
1
1 2 2
,
2 2 2
and the corresponding natural transformation
Γµ : Γt1,2,2,2u
Ñ Γt1,1,2,2u
r3 s
r2 s
x1 b x2 ÞÑ x1 x2 b x2 .
150
Chapter 8. Projective Generators
We now use the theorem to find out what happens to an element
y b xyr2s
P Γt1,2,2,2u pM q.
First, write the divided power as a multation
1
y
2
x
2
y
2
;
y
1
x
1 2
y y
then, simply compose the multations:
1
y
2 2
x y
Therefore,
2
y
1
1
1 2 2
2 2 2
Γµ py b xyr2s q xy b yr2s
2
y
1
2
y
1
y
2 2
.
x y
2pyr2s b xyq.
4
Like their cousins, the deviated powers, the divided power functors provide
a splitting of the fundamental functor into its atomic components.
Theorem 15.
Proof.
Γn HompBn , q à
„rns
|A|n
ΓA .
#A
Γn HompBn , M q Γn pM n q à
„rns
|A|n
#A
ΓA pM q.
§10. The Multi-Set Yoneda Correspondence
And finally, to close the chapter, yet another Yoneda Lemma. As usual, σ
denotes transportation maps.
Theorem 16: The Multi-Set Yoneda Lemma.
strict analytic functor. The map
Let A be a multi-set, and F a
ΥA,F : NatpΓA , F q Ñ FA: pB|#A q
ζ ÞÑ ζB#A p1brAs q
is an isomorphism of modules.
§10. The Multi-Set Yoneda Correspondence
151
The isomorphism is natural, in the sense that the following two diagrams commute:
pΓµ q
µ
ΥB,F
ΥA,F
NatpΓA , F q
/ F : pB|#A q
pξ:A qB|#A
NatpΓA , Gq
G
/ F : pB|#B q
B
A
ξ
ξ
/ F : pB|#A q
A
F rµs
σ
NatpΓB , F q
B
F
ΥA,F
NatpΓA , F q
A
ΥA,G
/ G: pB|#A q
A
Proof. A natural transformation
ζ : ΓA
ÑF
corresponds to a natural transformation
ηM : Γ|A| HompB#A , M q Ñ F pM q,
taking
#
0
αrX s ÞÑ
ζM
where αa : Ba
ηB#A
if X
A,
αp1qbrX s
if X A,
Ñ M, for a P #A. The original Yoneda map takes η to
¸
p1r|BA|s q ηB p1brX s q ηB p1brAs q ζB p1brAs q.
#A
„
| || |
#A
#A
#A
X #A
X
A
Thus, ΥA,F is the A-component of the original Yoneda map, and it is clear that
ΥA,F pζq ζB#A p1brAs q P FA: pB|#A q.
That the second diagram above commutes is clear, so there remains to
investigate the first. Evidently, since µ is a multation, A and B must be multisets of the same cardinality n. From the original Yoneda Lemma, there is a
commutative diagram:
NatpΓn HompB#A , q, F q
ΥB#A ,F
pσ qrµs NatpΓn HompB#B , q, F q
/ F pB#A q
F rµs
σ
ΥB#B ,F
/ F pB#B q
152
Chapter 8. Projective Generators
We shall show that the following diagram commutes, from which the claim
follows:
/ NatpΓn HompB#A , q, F q
NatpΓA , F q
pΓµ q
NatpΓB , F q
pσ qrµs / NatpΓn HompB#B , q, F q
Consider a natural transformation
ζ : ΓA
corresponding to an
Ñ F,
η : Γn HompB#A , q Ñ F
as above. The homomorphism
pσ qrµs
maps η to the natural transformation
ηM pσ qrµs : Γn HompB#B , M q Ñ F pM q,
Ñ M, takes
αrBs ÞÑ ηM pαrBs pσ qrµs q
ηM ppασqrµs q ζM ppασqp1qrµs q ζM pαp1qbrµs q.
which, for homomorphisms αb : Bb
On the other hand,
ζΓµ : ΓB
ÑF
corresponds to a natural transformation
Γn HompB#B , M q Ñ F pM q,
that takes
αrBs
ÞÑ ζM Γµ pαp1qbrBs q ζM pαp1qbrµs q.
The commutativity of the diagram is demonstrated.
Chapter 9
MODULE
REPRESENTATIONS
Et la glace où se fige un réel mouvement
Reste froide malgré son détestable ouvrage.
La force du miroir trompa plus d’un amant
Qui crut aimer sa belle et n’aima qu’un mirage.
Guillaume Apollinaire, La Force du Miroir
Exhibited on display in the preceding chapter were some rather special functors, most notably
BrHompBn , qsn
and Γn HompBn , q.
We rather vaguely suggested they were projective generators for the categories
Numn and Homn , respectively, postponing the proof to the present chapter.
After proving this, and invoking the celebrated Morita Equivalence, it can be
inferred that numerical and homogeneous functors of degree n may be viewed
as modules over the rings BrBnn sn and Γn pBnn q, respectively.
A first incarnation of this curious result, published in a Georgian journal
of some obscurity, appeared in 1988, when Professor Pirashvili, [17], established an equivalence of the category of polynomial functors (over Z) with a
module category. Fifteen years later, Dr. Salomonsson and Professor Ekedahl
discovered that strict polynomial functors likewise admit a module representation. This was based on previous work by Professors Friedlander and Suslin,
which only subsumed the special case of the base ring being a field. We refer
to Salomonsson’s doctoral thesis [20] for the general result.
We shall state and prove an equivalent version of Salomonsson’s theorem,
using the modern conveniences brought about by the invention of multations,
along with a modified version of Pirashvili’s, making it apply to numerical,
rather than polynomial, functors.
The formidable consequence of these two theorems is that polynomial
functors — of modules! — may in fact be viewed as modules themselves,
albeit over two diverent rings. Techniques of module theory may then enter
the game in order to study these functors.
153
154
Chapter 9. Module Representations
§1. The Divided Power Map
Presupposing for a moment the key rôle the rings BrBnn sn and Γn pBnn q are
about to play, one obvious way of relating our two species of functors — numerical and strict polynomial — is to find homomorphisms between the two
rings. We shall pave the way in this section, preparing for a full exploration
later.
Let us begin with somewhat greater generality. Start with a module M. We
propose a study of the divided power map
Ñ Γn pM q
x ÞÑ xrns .
γn : M
The divided power map is numerical of degree n and therefore induces
Theorem 1.
a linear map
γn : BrM sn
Ñ Γn pM q
rxs ÞÑ xrns .
This is a natural transformation of (numerical) functors.
Proof. Since γn is homogeneous of degree n, it is also numerical of the same
degree.
If x1 , . . . , xn
Lemma 1.
P M, then
px1 xn qrns x1 xn .
Proof. The Principle of Inclusion and Exclusion.
Theorem 2.
ism
is
Let M be finitely generated and free. The cokernel of the homomorph-
pπ, γn q : BrM sn Ñ BrM sn1 ` Γn pM q
Cokerpπ, γn q Γn pM q{ xx1 . . . xn | xi
In particular, pπ, γn q is an injection of finite index.
P My .
Proof. Let te1 , . . . , ek u be a basis for M. Then the elements
rf1 fm s,
P te1 , . . . , ek u,
for 0 ¤ m ¤ n, constitute a basis for BrM sn . The image of pπ, γn q is generated
fi
by the images
pπ, γn qprf1 fm sq rf1 fm s, rf1 fm srns
and
pπ, γn qprf1 fn sq 0, pf1 fn qrns
,
0 ¤ m n;
p0, f1 fn q.
§1. The Divided Power Map
155
The relations
rf1 fm s rf1 fm srns
mod Impπ, γn q
let us represent each element of the cokernel by a sum of divided nth powers,
while the relations
f1 . . . fn 0 mod Impπ, γn q
yield the desired factor module of Γn pM q.
Let M be finitely generated and free. The kernel of the homomorphism
Theorem 3.
γn : BrM sn
is
Ker γn
Ñ Γn pM q
Q bZ xrrzs rn rzs | r P B, z P M y X BrM sn .
Proof. Let te1 , . . . , ek u be a basis for M. Then the elements
rf1 fm s,
P te1 , . . . , ek u,
for 0 ¤ m ¤ n, constitute a basis for BrM sn .
Denote
fi
L Q bZ xrrzs r n rzs | r
then evidently
L X BrM sn
P B, z P M y ;
„ Ker γn .
We now show the reverse inclusion.
Calculating modulo L, we have
♦z
m
¸
I
(
„rms
m̧
p1q |I |
m
¸
P
z
i I
m̧
r 0
p1qmr
m
r
rrzs
" *
p1q m rn rzs m! n rzs,
m r
r 0
r
m
n
where m
denotes a Stirling number of the second kind.
We may then write
" *
n
m!
m
" *
rf1 fm s m! mn
¸
¸
I
„rms
p1qm|I |
¸
p1q |I | ♦m fi
iPI
I „rms
1
ξ ξ,
m
¸
P
i I
fi
156
Chapter 9. Module Representations
where ξ is a sum of mth deviations, and ξ1 collects the higher-order deviations.
We calculate ξ:
¸
ξ
I
„rms
¸
„rms
|A|m
#A
¸
♦ fa
P
a A
p1q |I | m
m
„ „rms
m
A
rms
|A|m
" *
m!
n
m
♦ fa
P
a A
♦ fa
P
A
#A I
#A
We thus have
m
A
# A„ I
|A|m
¸
¸
p1qm|I |
a A
m
1, . . . , 1
rf1 fm s m!rf1 fm s.
ξ1 mod L,
rf1 fm s m!rf1 fm s
and, consequently, provided 1 m n (so that
rf1 fm s m!
1
n
m
(
n
m
(
1
1 ξ
¡ 1),
mod L.
Now suppose ω P Ker γn . Using the above relation, together with
rs r0s 0 mod L
and
rf s rf s r0s n!
1
n
n
( ♦f
n
1
♦f
n! n
mod L,
we may express ω as a (fractional) linear combination of nth deviations of the
basis elements ei :
ω
¸
„rks
|A|n
cA
#A
Apply γn :
0 γn pωq ¸
„rks
|A|n
#A
♦ ea
P
a A
cA γn
mod L.
♦ ea
P
a A
¸
„rks
|A|n
cA eA .
#A
Because the elements erAs constitute a basis for Γn pM q, it must be that all coefficients cA 0, and hence ω P L. The proof is finished.
We remark that the divided power map may be extended, without much
ado, to a map
γn : Q bZ BrM sn Ñ Q bZ Γn pM q.
This map (almost trivially) possesses an inverse. We shall, however, be interested in obtaining an inverse under a slightly less generous localisation.
§1. The Divided Power Map
157
The homomorphism
Theorem 4.
εn : Γn pM q Ñ Q bZ BrM sn
xrAs ÞÑ
1
♦x
deg A A
constitutes a section of the divided power map:
1Γ pM q .
γn εn
n
This leads to a direct sum decomposition
Im εn
Proof. The relation γn εn
quence splits:
Γn pM q ` pKer γn X Im εn q.
1 is immediate, and then the following exact se-
/ Ker γn X Im εn
0
γn
/ Im εn o
εn
/
Γn pM q
/0
It will now be appropriate to specialise the preceding discussion to the
particular rings BrBnn sn and Γn pBnn q.
Theorem 5.
The maps
γn : BrBnn sn
Ñ Γn pBnn q,
εn : Γn pBnn q Ñ Q bZ BrBnn sn
are homomorphisms of algebras, when both rings are equipped with the product multiplication.
Proof. Calculate:
γn prxsq γn prysq xrns yrns
pxyqrns γn prxysq γprxs rysq.
To show εn preserves multiplication, it will be enough to consider pure
powers xrns and yrns . Use the Deviation Formula:
1
εn pxrns q εn pyrns q ♦x
1
♦y
n! n
n! n
1
pn!q2 n! ♦n xy
1
♦ xy εn ppxyqrns q εn pxrns yrns q.
n! n
158
Chapter 9. Module Representations
§2. Module Functors
Lemma 2.
A module functor that vanishes on B8 is identically zero.
Proof. Let
F : XMod Ñ Mod
be a module functor. Since F is extended to arbitrary modules through inductive limits, we have by definition
n
F pB8 q lim
ÝÑ F pB q.
Note that F pM q is always a direct summand in F pM ` N q, for modules M
and N. This follows from the Cross-Evect Decomposition:
F pM ` N q loooooomoooooon
F : pq ` F : pM q `F : pN q ` F : pM | N q.
p q
F M
Since Bn is always a direct summand of B8 , any F pBn q is a direct summand
of F pB8 q.
Theorem 6.
The fundamental module functor
BrHompB8 , qs
is a small projective generator for
FunpXMod, Modq,
Num8
through which there is a Morita equivalence
FunpXMod, Modq BrB8ℵ0 s Mod,
where BrB8ℵ0 s carries the product multiplication.
More precisely, the functor F corresponds to the abelian group F pB8 q, with module
structure given by the equation
rτs x F pτqpxq.
Proof. Step 1: BrHompB8 , qs is projective. We must show that
NatpBrHompB8 , qs, q
is right-exact, or preserves epimorphisms. Hence let η : F Ñ G be epic, so that
each ηM is onto. The following diagram, constructed by aid of the Yoneda
Lemma, shows that η is epic:
NatpBrHompB8 , qs, F q o
η
Υ
ηB8
NatpBrHompB8 , qs, Gq o
/ F pB8 q
Υ
/ GpB8 q
§2. Module Functors
159
Step 2: BrHompB8 , qs is a generator. By the lemma,
0 NatpBrHompB8 , qs, F q F pB8 q
implies F 0.
Step 3: BrHompB8 , qs is small. Compute, using the Yoneda Lemma:
Nat BrHompB8 , qs,
à
à
à
Fk pB8 q Fk pB8 q
à
Nat BrHompB8 , qs, Fk .
As FunpXMod, Modq is an abelian category
Fk
Step 4: The Morita equivalence.
with arbitrary direct sums, we have a Morita equivalence:
Nat B Hom B8 ,
p r
FunpXMod, Modq
l
p
B Hom B8 ,
r
p
qs,q
+
S Mod
qsbS The new base ring is
S
Nat BrHompB8 , qs
BrEnd B8 s BrB8ℵ s.
0
Plainly, the functor F corresponds to the abelian group
NatpBrHompB8 , qs, F q F pB8 q.
Step 5: The module structure. Under the Yoneda map, an element x
will correspond to the natural transformation
P F pB8 q
ηM : BrHompB8 , M qs Ñ F pM q
rαs ÞÑ F pαqpxq.
Likewise, a scalar rτs P BrB8ℵ0 s will correspond to
σM : BrHompB8 , M qs Ñ BrHompB8 , M qs
rαs ÞÑ rα τs.
The product of the scalar σ and the module element η is the transformation
pη σqM : BrHompB8 , M qs Ñ F pM q
rαs ÞÑ F pα τqpxq,
which under the Yoneda map corresponds to
pη σqB8 pr1B8 sq F p1B8 τqpxq F pτqpxq P F pB8 q.
The scalar multiplication on F pB8 q is thus given by the formula
rτs x F pτqpxq,
and the proof is finished.
160
Chapter 9. Module Representations
Example 1.
The functor T 2 corresponds to the abelian group
T 2 pB8 q B8 b B8 ,
with module structure
rτs px b yq T 2 pτqpx b yq τpxq b τpyq,
for any rτs P BrB8ℵ s.
4
0
§3. Numerical Functors
We now repeat the feat, but for numerical functors.
A polynomial functor of degree n that vanishes on Bn is identically zero.
Lemma 3.
Proof. Suppose that F is polynomial of degree n, and that F pBn q 0. We shall
show that F pBq q 0 for all natural numbers q.
Consider first the case q ¤ n. Then Bq is a direct summand of Bn , so F pBq q
is a direct summand of F pBn q 0.
Proceed by induction, and suppose F pBq1 q 0 for some q 1 ¥ n. Decompose
1Bq π1 πq ,
where πj : Bq Ñ Bq as usual denotes the jth projection. Since F is polynomial
of degree n, and therefore of degree q 1,
0 F pπ1 πq q J
Consider a J with |J |
morphism F
°
J
πj
¤ q 1.
Since
¸
„rqs
°
J
p1qq|J | F ¸
πj .
J
πj factors through Bq1 , the homo-
factors through F pBq1 q
0.
Only J
rqs will give a
non-trivial contribution to the sum above, yielding
hence F pBq q 0.
Theorem 7.
0 F pπ1
πq q F p1Bq q 1F pBq q ;
The fundamental numerical functor
BrHompBn , qsn
is a small projective generator for Numn , through which there is a Morita equivalence
Numn
BrB s Mod,
n n n
where Br
sn carries the product multiplication.
More precisely, the functor F corresponds to the abelian group F pBn q, with module
structure given by the equation
Bn n
rτs x F pτqpxq.
§4. Homogeneous Functors
161
Proof. The preceding proof goes through exactly as before, but with Lemma 3
in place of Lemma 2.
Let n 2. The functor T 2 corresponds to the abelian group
Example 2.
T 2 pB2 q B2 b B2 ,
with module structure
rτs px b yq T 2 pτqpx b yq τpxq b τpyq,
for any rτs P BrB22 s2 .
This should be compared with the module obtained in Example 1. Apparently this latter module contains vast amounts of superfluous data, and may
be cut down in size considerably, once we take advantage of the fact that the
functor is quadratic.
4
§4. Homogeneous Functors
The fundamental homogeneous functor
Theorem 8.
Γn HompBn , q
is a small projective generator for Homn , through which there is a Morita equivalence
Homn
Γ pB q Mod,
n
n n
where Γn pBnn q carries the product multiplication.
More precisely, the functor F corresponds to the abelian group F pBn q, with module
structure given by the equation
τrns x F pτqpxq.
Proof. The previous proof goes through exactly as before.
Example 3.
abelian group
Let n
2. The functor T 2 corresponds once again to the
T 2 pB2 q B2 b B2 ,
with module structure
τr2s px b yq T 2 pτqpx b yq τpxq b τpyq,
for any τr2s
P Γ2 pB22 q.
4
162
Chapter 9. Module Representations
§5. Quasi-Homogeneous Functors
With all the ground-work laid in the introductory section, the module-theoretic interpretation of quasi-homogeneity is immediate.
Theorem 9.
Let F be a numerical functor, corresponding to the BrBnn sn -module
M. The functor F is quasi-homogeneous of degree n iff M is a module over
Im γn
BrBnn sn { Ker γn .
Proof. Recall that the scalar multiplication of BrBnn sn on M
by
rσsx F pσqpxq, σ P Bnn , x P F pBn q.
F pBn q is given
The requirement that Ker γn annihilate F pBn q is equivalent to demanding that
F itself vanish on
Ker γn
@
D
Q bZ rrσs rn rσs | r P B, σ P Bnn X BrBnn sn ,
which clearly would be a consequence of quasi-homogeneity.
To show that, conversely,
σ P Bnn ,
F prσq r n F pσq,
implies quasi-homogeneity, we first show that
F pr 1Bq q r n F p1Bq q
for all natural numbers q. This is clear when q ¤ n, for then 1Bq factors through
1Bn . When q ¡ n, split up into the canonical projections, and use induction:
F pr 1Bq q F prπ1
¸
I
€rqs
¸
I
€rqs
rπq q
p1q |I | F
q
¸
P
rπi
i I
p1q |I | rn F
q
rn F pπ1 ¸
P
πi
i I
πq q r n F p1Bq q.
Ñ Bq , we have
F prαq F pr 1B qF pαq r n F p1B qF pαq r n F pαq.
Finally, for an arbitrary homomorphism α : Bp
q
q
Chapter 10
COMBINATORIAL
REPRESENTATIONS
“Here is Ts’ui Pên’s labyrinth,” he said, indicating a tall lacquered desk.
“An ivory labyrinth!” I exclaimed. “A minimum labyrinth.”
“A labyrinth of symbols,” he corrected. “An invisible labyrinth of time. [. . . ]”
Jorge Luis Borges, The Garden of Forking Paths
At the beginning of the twenty-first century, the team Baues, Dreckmann,
Franjou & Pirashvili sought methods to combinatorially encode Z-module
functors, with a particular eye on polynomial ones. Their design was to build
a clever two-way correspondence (category equivalence) between functors
N XMod
Ñ Z Mod,
on the one hand, and Mackey functors (whatever that may be) from the category
Sur, on the other. This marvellous theorem, the first of its kind, enabled the
reduction of a functor to a significantly smaller collection of data.
Passing to polynomial functors has the evect of erasing the distinction
between N-modules and Z-modules, and the quartet’s main result may then
be stated, without resorting to Mackey functors: Polynomial (Z-)functors of degree
y n.
n are equivalent to linear (additive) functors from Sur
They left unresolved the interesting question of how to systematise the full
category of Z-module functors. Moreover, attempts to generalise their argument to arbitrary base rings will encounter diHculties, as it is not clear what
category should play the rôle of Sur. Here the labyrinth category comes to
the rescue, constituting, as it does, the universal means of combinatorially encoding all module functors (not just polynomial ones) over any (unital) base
ring (not just commutative ones). The generalisation to arbitrary rings reads:
Numerical functors of degree n are equivalent to linear functors from Labyn . This includes the case n 8, corresponding to general module functors. To connect
with the above result, recall that we have previously exhibited a category isomorphism
y n Laby .
Sur
Z
n
After the quartet had announced the publication of their results in [1], Torsten Ekedahl and Pelle Salomonsson quickly followed suite. They showed in
163
164
Chapter 10. Combinatorial Representations
[20] how strict polynomial functors (over any commutative base ring) may be
similarly encoded, this time as Mackey functors from the category of multisets and multijections. Striving, as we do, to avoid the language of Mackey
functors at any cost, we have chosen to replace multijections with multations.
The pair’s theorem, which we reprove in this setting, then asserts the following: Homogeneous functors of degree n are equivalent to linear functors from MSetn .
§1. Module Functors
As is our custom when dealing with general module functors, we suppose the
base ring ring B unital only, and not necessarily commutative. As before, free
and finitely generated modules are automatically bimodules, but all homomorphisms are right module homomorphisms.
Two kinds of functors will be considered. Module functors XMod Ñ Mod
may, of course, be of an arbitrary nature (linear, polynomial, numerical, and
what not). On the other hand, labyrinth modules Laby Ñ Mod shall always be
assumed B-linear.
As usual, σ denotes transportation maps.
Definition 1.
Given a homomorphism
α
¸
P P
sba σba : BA
Ñ BB ,
a A, b B
(a B A matrix) its associated maze S : A Ñ B is
S
!
a
sba
/b
a P A, b P B
)
.
Associated mazes are always simple. Note that, if but a single component
sba vanish, the associated maze S 0.
Theorem 1.
1. Let P be the associated maze of α : BA Ñ BB , and Q the associated maze of
β : BA Ñ BB . Then the associated maze of α β is P ` Q.
2. Let P be the associated maze of α : BB Ñ BC , and Q the associated maze of
β : BA Ñ BB . Then the associated maze of α β is P e Q.
This theorem should retroactively motivate our interest in the operations
` and e, as well as our choice of notation.
Getting down to business, we wish now to define a functor
Φ : FunpXMod, Modq Ñ FunpLaby, Modq,
§1. Module Functors
165
which will eventually turn out to be an equivalence. Given a module functor
F : XMod Ñ Mod, the corresponding labyrinth functor should take a finite set
to the corresponding cross-evect:
X
ÞÑ F : pB|X q.
Mazes should be interpreted as deviations, in the following sense:
rP : X Ñ Y s ÞÑ F
pσyx ♦
rp : xÑysPP
F: B X
p | qÑF : pB|Y q
.
Restricting action to the appropriate cross-evects is in fact an unnecessary
caution, as shown by the following lemma.
The map
Lemma 1.
F
♦
rp : xÑysPP
is in fact a map
pσyx
: F pBX q Ñ F pBY q
F : pB|X q Ñ F : pB|Y q,
in the sense that all other components vanish.
Proof. We use Theorem 7.4. If τx is any insertion with x
and hence
F
♦
rp : xÑysPP
Similarly,
F pρy qF
F pτx q 0.
pσyx
♦
rp : xÑysPP
P X, then σyx τx 0,
pσyx
0,
when ρy is any retraction with y P Y .
For the functor
ΦpF q : Laby Ñ Mod,
we thus propose the following definition:
rP : X Ñ Y s ÞÑ
Lemma 2.
F
X
ÞÑ F : pB|X q
♦
rp : xÑysPP
pσyx
ΦpF q is a functor Laby Ñ Mod.
: F : pB|X q Ñ F : pB|Y q .
166
Chapter 10. Combinatorial Representations
Proof. That ΦpF q respects the relations in Laby follows from
ΦpF q P Y t x
0
/yu
F p 0q 0,
and
ΦpF q P Y t x
/yu
F p pa bqσyx q
F p aσyx q F p bσyx q F p aσyx bσyx q
b /
a /
y
y
u ΦpF q P Y t x
u
ΦpF q P Y t x
a b
ΦpF q P Y t x
a
b
/
/y
u
.
Functoriality of ΦpF q follows from the Deviation Formula and the definition
of maze composition.
Let ζ : F
Ñ G be a natural transformation. Define
Φpζq : ΦpF q Ñ ΦpGq
by restriction to the appropriate cross-evects:
ΦpζqX
Lemma 3.
ζ:B|
X
: F : pB|X q Ñ G: pB|X q.
Φ is a functor
FunpXMod, Modq Ñ FunpLaby, Modq.
Proof. By the Labyrinthine Yoneda Lemma,
ζ:B| : F : pB|X q Ñ G: pB|X q
X
corresponds to
ζ : Natp∆X , F q Ñ Natp∆X , Gq.
We now construct the inverse of Φ. Let
H : Laby Ñ Mod
be a functor. Define
by
Φ1 pH q : XMod Ñ Mod
Φ1 pH qpBA q à
„
X A
H pX q.
§1. Module Functors
167
Also, let
α
¸
P
P
sba σba : BA
Ñ BB
a A
b B
be a homomorphism, and let S be its associated maze. Define the component
H pX q Ñ H pY q of Φ1 pH qpαq to be
¸

P „S 
X
H pP q.
ÑY
This amounts to saying that
Φ1 pH qpαq ¸
¸
H pP q.
„ „
„
X A P S
Y B
X ÑY
Note in particular that

Φ1 pH qpSqH pY qÑH pZq
if
Y
but
∅Z
or Y

Φ1 pH qpSqH p∅qÑH p∅q
0
∅ Z,
H pI∅ q IH p∅q .
Φ1 pH q is a module functor.
Lemma 4.
Proof. Let
α
¸
P
P
scb σcb : BB
Ñ BC
and
β
¸
tba σba : BA
P
P
Ñ BB
a A
b B
b B
c C
be homomorphisms, and let S and T be their associated mazes. The associated
maze of α β is then S e T .
Let X „ A and Z „ C. By Theorem 3.2,
¸

Φ1 pH qpS T qH pX qÑH pZq H pW q

W „pSeT q
¸
X

W „pSbT q
X
ÑZ
H pW q,
ÑZ
while

Φ1 pH qpSq Φ1 pH qpT q H pX qÑH pZq
¸ 

Φ1 pH qpSq
Φ1 pH qpT q
„
Y B
p qÑH pZq
H Y
p qÑH pY q
H X
168
Chapter 10. Combinatorial Representations
¸ „
Y B
¸
¸

P „S 
¸
Y
H pP q
¸
ÑZ ¸

Q „T 
X
„ „  Q„ T 

Y ÑZ
X ÑY
H pPQq
Y B P S
¸
¸

Y „ B P „S 
Y
¸
V
¸
ÑZ
„pSbT qX ÑZ

Q„ T 
ÑY
¸
H
X
H pQq
„b
V
V P Q
ÑY
H pV q.


The last step comes from noting that every submaze of pS bT qX ÑZ is obtained
as V „ P b Q, for some P and Q. The functoriality of Φ1 pH q follows.
Lemma 5.
ΦpΦ1 pH qq H.
Proof. Let P : X
Ñ Y be a maze, and calculate the deviation
Φ1 pH q
♦
rp : xÑysPP
pσyx
¸
„
p1q|P||S| Φ1 pH q ¸
P
pσyx .
(1)
p S
S P
The sum should run over all sub-multi-sets S of P (rather than just submazes).
Beginning with
Φ1 pH q ¸
P
pσyx ,
p S
the component H pZ1 q Ñ H pZ2 q is
¸

Q„ S 
Z1
H pQq.
ÑZ2
The component H pZ1 q Ñ H pZ2 q of (1) is then
¸
„
p1q|P||S|
S P
¸

Q„ S 
Z1
H pQq ÑZ2
¸

Q „P 
Z1
p1q|P| H pQq
ÑZ2
¸
„„
p1q|S| .
Q S P
The inner sum vanishes
if P Q, and it equals p1q|P| if P Q. In thelatter

case, since Q „ P Z1 ÑZ2 , it must be that Z1 X, Z2 Y , and Q P P X ÑY .
Consequently,
Φ1 pH q
♦
pσyx rp : xÑysPP
p qÑH pZ2 q
H Z1
#
H pP q if Z1 X and Z2
0
else.
Y,
§1. Module Functors
169
It is then an immediate consequence that
ΦpΦ1 pH qqpX q Φ1 pH q: pB|X q Im Φ1 pH q
♦ πx
Im H pIX q Im 1H pX q H pX q,
and also that
ΦpΦ1 pH qqpP q Φ1 pH q
Lemma 6.
♦
rp : xÑysPP
pσyx
P
x X
H pPq.
Φ1 pΦpF qq F.
Proof. We have, by the Cross-Evect Decomposition,
Φ1 pΦpF qqpBA q à
„
X A
ΦpF qpX q à
„
X A
F : pB|X q F pAq.
Let a homomorphism
α
¸
Ñ BB
sba σba : BA
P P
a A, b B
be given, with associated maze S. Then
Φ1 pΦpF qqpαq ¸
¸
ΦpF qpP q
„ „
„
X A P S
Y B
X ÑY
¸
¸
F
„ „
„
X A P S
Y B
X ÑY
¸
¸
♦
rp : xÑysPP
X A P S
Y B
X ÑY
¸
„
F
E A B
F
♦
px,yqPE
¸
px,yqPAB
pσyx
F
„ „
„
♦
rxÑysPP
syx σyx
syx σyx
syx σyx F pαq.
Assembling these results, we obtain the following marvellous theorem.
(The functoriality of Φ1 is a direct consequence of the lemmata, and need
not be established separately.)
170
Chapter 10. Combinatorial Representations
The functor
Theorem 2: Labyrinth of Fun.
ΦLaby : FunpXMod, Modq Ñ FunpLaby, Modq,
where ΦLaby pF q : Laby Ñ Mod takes
rP : X Ñ Y s ÞÑ
F
X
ÞÑ F : pB|X q
♦
rp : xÑysPP
pσyx
: F : pB|X q Ñ F : pB|Y q ,
is an equivalence of categories.
§2. Polynomial Functors
Since mazes correspond to deviations, the following simple characterisation
of polynomiality should come as no surprise.
Theorem 3.
The module functor F is polynomial of degree n iff ΦLaby pF q vanishes
on sets with more than n elements.
Proof. Assume first that F is polynomial of degree n. Since mazes with k passages correspond to kth deviations, ΦLaby pF q will certainly vanish on mazes
with more than n passages.
Suppose now, conversely, that ΦLaby pF q vanishes on mazes with more than
n passages. Consider n 1 homomorphisms
α1 , . . . , αn
BA
1:
with associated mazes
P1 , . . . , Pn
1:
Ñ BB ,
A Ñ B,
respectively. These mazes are all similar, and we may label consistently the
passages of each Pi by
pi1 , . . . , pim .
Let X „ A and Y
Note that if
„ B be sets.
pij | j
PJ
(
is a legitimate submaze of Pi for one particular i, it is so for all choices of i.
When this is the case, we say that the set J „ rms is admissible. Then also
#
¸
P
pij j
PJ
+
i I
is a legitimate submaze of
ð
P
i I
Pi X
ÑY
ð
P
i I

Pi X ÑY
§3. Numerical Functors
171
for any I „ rn 1s. Note that this is the associated maze of the sum
We are now ready to calculate the deviation of F:

F pα1 αn 1 qF : pB| qÑF : pB| q
X
Y 

¸
¸
p1qn 1|I | F αi 
:
:
I
I
„rn 1s
¸
I
|I |
„rn 1s
p1q
Q
p1q
J
P
i I
¸
|I |
n 1
„rn 1s
Ð
„
„rms
Pi
P αi .
i I
pB|X qÑF pB|Y q
F
¸
n 1
¸
P
i I
°
ΦLaby pF qpQq


X
Ñ#
Y
ΦLaby pF q
¸
P
pij j
PJ
+
,
i I
where the inner sum is taken over admissible J only. For a set K „ I J, let KI
and KJ denote the projections on I and J, respectively. We may use Theorem
3.3 to transform the latter sum to

F pα1 αn 1 qF : pB| qÑF : pB| q
¸
I
„rn 1s
p1qn
K
K
X
¸
|I |
J
¸
1
„rn 1srms
¸
„rn 1srms
KI rn 1s
„ „rn 1s
KI I
rn
ΦLaby pF qp pij
p1qn
¸
|I | 1
(
pi, jq P K q
(
ΦLaby pF qp pij
pi, jq P K q
J KJ
ΦLaby pF qp pij
The condition KI
Y
„rms K „I J
K J J
¸
¸
(
pi, jq P K q.
1s implies |K | ¥ n
pij
1, and so all mazes
pi, jq P K
(
will contain more than n passages. The sum will therefore equal 0, by the
hypothesis on ΦLaby pF q.
§3. Numerical Functors
We now investigate how to interpret numericality in the labyrinthine setting.
The base ring B is assumed numerical.
Lemma 7.
Then
Let r
q
¹
j 1
P B, and let n, w1 , . . . , wq be natural numbers satisfying
w1 wq ¤ n.
r
wj
ņ
m 0
m̧
r
m
q ¹ k
p1q m
m k
k 0
k
j 1
wj
.
172
Chapter 10. Combinatorial Representations
Proof. We prove the formula when r is an integer, and then refer to the Numerical Transfer Principle.
Fix a subset S „ rr s with |S| m. Suppose
we wish to choose subsets
”
W1 , . . . , Wq „ S, such that |Wi | wi and
Wi S. By the Principle of
Inclusion and Exclusion, this can be done in
m̧
p1qmk
q m ¹ k
k
k 0
j 1
wj
ways. The quantites
q
¹
j 1
r
wj
8̧
m̧
m 0
r
m
q ¹ k
p1q m
m k
k
k 0
j 1
wj
will then both count the total number of ways to choose subsets W1 , . . . , Wq
rrs, such that |Wi | wi .
„
Theorem 4.
The module functor F is numerical of degree n iff ΦLaby pF q factors
through Labyn . The functor ΦLaby induces an equivalence of categories
Numn
Ñ FunpLabyn , Modq.
Proof. Recall from Chapter 3 that, when A is a multi-set supported in the maze
P, we let EA denote the maze
EA
!
¤
rp : xÑysPA
x
1
/y
)
,
with all passages of A reassigned the label 1.
The theorem states that the functor F is numerical of degree n iv the equation
¸ ¹
p
ΦLaby pF qpP q ΦLaby pF qpEA q
degA p
#AP pPP
|A|¤n
holds for all mazes P. It should be clear from Theorem 6.4 that numerical
functors satisfy this equation.
Suppose now, conversely, that ΦLaby pF q satisfies the equation. It will then
certainly vanish on mazes with more than n elements, whence F is polynomial
of degree n. We now wish to use Theorem 6.3, and thus seek to evaluate
F pr 1Bn q The component
¸
„d
P r Irns
ΦLaby pF qpP q.
ΦLaby pF qpX q Ñ ΦLaby pF qpY q
§3. Numerical Functors
173
of this is 0 if X Y . Turning to the case X Y , we may without loss of
generality assume X Y rqs. Then the component
ΦLaby pF qprqsq Ñ ΦLaby pF qprqsq
is
ΦLaby pF qpr d Irqs q q
¹
¸
rqs j1
|A|¤n
#A
where we let wj
of
r
F ♦ 1Bn
m
m
m 0
is
m̧
ņ
r
m
m 0
¤ wq n j 1
r
ΦLaby pF qpEw q,
wj
degA j ¥ 1. Similarly, the component
ΦLaby pF qprqsq Ñ ΦLaby pF qprqsq
ņ
q
¹
¸
w1
r
ΦLaby pF qpEA q
degA j
p1qmk
k 0
m̧
ņ
r
m
m 0
m
k
k
k 0
q
¹
¸
w1
p1q m F pk 1Bn q
m k
¤ wq n j 1
k
ΦLaby pF qpEw q.
wj
It is now only a matter of using the lemma, to establish the equality
F pr 1Bn q ņ
r
F ♦ 1Bn
m
m
m 0
.
Consequently, F is numerical.
Example 1.
Let us take a simple example. If the labyrinth module H corresponds to a module functor of degree 3, it will satisfy the equation
8
a b
qqq
H
H MM M&
1
1
a
b
1
8 qqqqqq8
b
H MM1 M
1
1 &
a
2
q8 qMqMM& 1
1
and also
H
a
b
E a
1
b
H
1
1
1
E a
1
1 8 b
qMqq1 H M MM
,
MM &
2
1 & 174
Chapter 10. Combinatorial Representations
a
2
b
H
1
1
1
' E a
1
b
H
2
1
1
1
7 E .
1
4
We now pay homage to our predecessors. The main result of Professor
Pirashvili’s team ([1]) is the following1 : Polynomial functors on the category of Zy n to Z-modules. For the ring Z, the
modules are equivalent to linear functors from Sur
concepts “polynomial” and “numerical” coincide; hence this theorem may be
written as
y
Z Numn FunpSurn , Z Modq.
It is an immediate consequence of the isomorphisms
yn
Sur
Z Labyn
and Numn
FunpLabyn , Modq.
§4. Quasi-Homogeneous Functors
Theorem 5.
The module functor F is quasi-homogeneous of degree n iff ΦLaby pF q
factors through Labyn . The functor ΦLaby induces an equivalence of categories
Ñ FunpLabyn , Modq.
QHomn
Proof. Let F be quasi-homogeneous, and let a P Q bZ B. For any deviation, we
have
F paα1 aαk q an F pα1 αk q,
and we may calculate for a pure maze P:
an ΦLaby pF qpP q an F
♦
rxÑysPP
σyx
F
♦
rxÑysPP
aσyx
ΦLaby pF qpa d Pq.
Conversely, assume ΦLaby pF q factors via Labyn . Then, for any k P N,
an F p1Bk q an
¸
K
¸
K
„rks
„rks
ΦLaby pF qpIK q
ΦLaby pF qpa d IK q F pa 1Bk q.
1 We point out that they restrict their attention to pointed functors. We have circumvented this
restriction by including ∅ among the finite sets.
§5. Quadratic Functors
175
According to a structure theorem for the labyrinth categories, there are
isomorphisms
B Labyn
B bZ Z Labyn
n
B Laby
and
B bZ Z Labyn .
Their significance may be summarised thus. The category of numerical (or
quasi-homogeneous) functors over an arbitrary numerical ring is identical
in structure to the category of numerical (respectively, quasi-homogeneous)
functors over Z. These are the equations to prove this:
B Numn
FunB pB Labyn , B Modq
FunB pB bZ Z Labyn , B Modq
FunZ pZ Labyn , B Modq.
The corresponding result for homogeneous functors holds more trivially.
By definition, for any ring B,
B MSetn
which leads to
B Homn
B bZ Z MSetn ,
FunZ pZ MSetn , B Modq.
§5. Quadratic Functors
A few examples of labyrinth representations are in order. Let us take rns as
the canonical representative of sets of cardinality n.
Example 2.
Let C pBn q
ΦLaby pC q will take
K be a constant functor.
r0s ÞÑ K,
The labyrinth functor
r1s, r2s, r3s, . . . ÞÑ 0.
4
Example 3.
Let F pBn q K ` Ln be an aHne functor. ΦLaby pF q will take
r0s ÞÑ K,
and map the maze
r1s ÞÑ L,
1
c
/1
r2s, r3s, . . . ÞÑ 0,
ÞÑ rc : L Ñ Ls .
4
Let us now determine the structure of Num2 by classifying the quadratic
numerical functors. The key point is unravelling the structure of the category
Laby2 . It contains three non-isomorphic objects: r0s, r1s, and r2s. We observe
the following relations:
a
/
a
1
1
/
a
2
1
1
/
/
176
Chapter 10. Combinatorial Representations
A
–
C
–
A
A
B
C
S
B
I
C
2A
–
2C
–
S
–
2B
–
S
–
B
–
I
Table 10.1: Multiplication table for Laby2 .
and
a
b
/ 1
1
b
/
a
/
1
1
/
.
Consequently, every maze in Laby2 can be reduced to (linear combinations of)
identity mazes and the following:
A
C
1 LL / 1
L&
1 2
1
1
1
1
/1
/
B
S
1
1 r/8 1
rr 1
2
1 MMq1 8 1
qqM&
2 12
The (skeletal) structure of the category Laby2 is thus reduced to the following,
promptly suggesting the nickname dogegory:
)
I
C
r0s
3 r1s
A
d
$
S
r2S s s
B
I
I
The mazes A, B, C, and S are not independent. Their multiplication table is
given in Table 10.1. Clearly we can do with only A, B and S, and we obtain the
following explicit description of Num2 .
Theorem 6.
A quadratic numerical functor is equivalent to the collection of modules K, X, and Y ; together with homomorphisms α, β, σ as indicated, subject to the
following four relations:
αβ 1
βσ β,
σ,
σα α,
α
K
Xa
!
Y f
σ2
1.
σ
β
The reader will no doubt note that we can, in fact, also dispense with
σ αβ 1, and let the two homomorphisms α and β be subject to a meagre
two relations:
βαβ 2β,
αβα 2α.
§5. Quadratic Functors
177
We now describe the four classical quadratic functors. Because they are
of the second degree, and because they are all reduced, the module K 0.
We will juggle with two isomorphic copies of B, denoted by B1 xe1 y and
B2 xe2 y.
The functor ΦLaby pT 2 q has modules
Example 4.
X
pT 2 q: pB1 q xe1 b e1 y ,
Y
pT 2 q: pB1 |B2 q xe1 b e2 , e2 b e1 y
and homomorphisms
e1 b e1 ÞÑ e1 b e2 e2 b e1
e1 b e2 , e2 b e1 ÞÑ e1 b e1
e1 b e2 ÞÑ e2 b e1
e2 b e1 ÞÑ e1 b e2 .
α:
β:
σ:
4
The functor ΦLaby pS2 q has modules
Example 5.
X
pS2 q: pB1 q xe12 y ,
Y
pS2 q: pB1 |B2 q xe1 e2 y
and homomorphisms
e12 ÞÑ 2e1 e2
e1 e2 ÞÑ e12
e1 e2 ÞÑ e1 e2 .
α:
β:
σ:
4
The functor ΦLaby p
Example 6.
Λ2
X pΛ q: pB1 q xe1 ^ e1 y 0,
2
q has modules
Y pΛ2 q: pB1 |B2 q xe1 ^ e2 y
and homomorphisms
α:
β:
σ:
0
0
e1 ^ e2
ÞÑ e1 ^ e2 .
4
The functor ΦLaby pΓ2 q has modules
Example 7.
X
pΓ2 q: pB1 q A
r2 s E ,
e1
Y
pΓ2 q: pB1 |B2 q xe1 e2 y
and homomorphisms
α:
β:
σ:
r2s ÞÑ e e
1 2
r2s
e e ÞÑ 2e
e1
1 2
e1 e2
1
ÞÑ e1 e2 .
4
178
Chapter 10. Combinatorial Representations
§6. Labyrinth Modules
Our description of the category equivalence
FunpLabyn , Modq,
Numn
explicit though it be, has the disadvantage of obscuring the fact that this is
another instance of Morita equivalence. A shorter, possibly more conceptual
proof, would run as follows.
Let n P Z Y t8u. The decomposition
BrHompBn , qs à
X
„rns
∆X .
shows that the functors ∆X constitute a family of small projective generators
for Numn . Letting
L : Labyn Ñ Devn Ñ Numn
denote the contra-variant inclusion functor, there is a Morita equivalence:
p q
Nat L,
)
FunpLabyn , Modq
Numn i
b
L Labyn
As an immediate consequence of this latter view, the labyrinth functor
H : Labyn Ñ Mod corresponds to the module functor F L bLabyn H, given
by
pL bLaby H qpM q à
n
P
X Labyn
à
P
X Labyn
LpX qpM q b H pX q
∆X pM q b H pX q.
Imposed upon this module are the following relations:
b H pPqpzq UP b z,
for any z P H pX q, U P ∆Y pM q, and maze P : X Ñ Y . Interpreting this in terms
U
of F, we have the following theorem, where σ denotes the usual canonical
transportation maps.
Theorem 7.
Let F be a numerical functor of degree n, where n
module F pM q is the quotient of
à
P
X Labyn
by all relations
U
bF
∆X pM q b F : pB|X q
♦
rp : xÑysPP
pσyx
pzq UP b z,
P Z Y t8u. The
§7. Homogeneous Functors
179
for any z P F : pB|X q, U P ∆Y pM q, and maze P : X Ñ Y .
Moreover, when α : M Ñ N is a homomorphism,
F pαq à
∆X pαq b 1F : pB|X q .
P
X Labyn
We will often abuse notation, and write simply
F pM q à
P
X Labyn
∆X pM q b F : pB|X q,
with the imposed relations tacitly understood.
The advantage of this viewpoint will become evident as soon as we try to
compute the composition of two module functors F and G (which will be used
later when considering the plethysm). We may write
F pGpM qq à
P
X Labyn
∆X pGpM qq b F : pB|X q,
and we obtain the following formula for the deviations of a composition.
Theorem 8.
pF Gqpα1 αm q ¸
P
¸
r s
¤
"
P
p
G ♦iPI αi
q /
*
b 1F : pB|
X
I J
X Labyn J C m
q.
Proof. Calculate, according to Theorems 7.3 and 8.7:
pF Gqpα1 αm q ¸
P
X Labyn
¸
P
∆X pGpα1 αm qq b 1F : pB|X q
¸
r s
X Labyn J C m
¸
P
¸
r s
X Labyn J C m
∆X
♦G
P
I J
¤
P
"
♦ αi
P
i I
p
G ♦iPI αi
I J
b 1F : pB|
X
q /
*
q
b 1F : pB|
X
q.
§7. Homogeneous Functors
We now turn to combinatorially interpreting homogeneous polynomial functors, and cite [20] as our reference. Specifically, we do not obtain any new
results in this section — they were known previously to at least two people
— but simply rephrase old ones in the smooth language of multations. This
allows for, we believe, the most economic formulation of the theory.
180
Chapter 10. Combinatorial Representations
Let B be commutative and unital. We propose to establish an equivalence
of categories
Homn FunpMSetn , Modq,
and we emphasise, as we did in the labyrinth case, that functors MSetn Ñ Mod,
shall always be assumed linear. These functors go by the name of Schur modules.
Let F be a homogeneous functor of degree n. Define
ΦpF q : MSetn
Ñ Mod
by the formulæ
rµ : A Ñ Bs ÞÑ
Lemma 8.
A ÞÑ FA: pB|#A q
Fσrµs : FA: pB|#A q Ñ FB: pB|#B q .
ΦpF q is a functor MSetn
Ñ Mod.
Proof. Functoriality is clear from the Multi-Set Yoneda Lemma, as Fσrµs corresponds to
pΓµ q : NatpΓA , F q Ñ NatpΓB , F q.
Let ζ : F
Ñ G be a natural transformation. Define
Φpζq : ΦpF q Ñ ΦpGq
by restriction to the appropriate multi-cross-evects:
ΦpζqA
Lemma 9.
pζ:A qB|
#A
: FA: pB|#A q Ñ G:A pB|#A q.
Φ is a functor
Homn
Ñ FunpMSetn , Modq.
Proof. By the Multi-Set Yoneda Lemma, pζ:A qB|#A corresponds to
ζ : NatpΓA , F q Ñ NatpΓA , Gq.
We now construct the inverse of Φ. Let
J : MSetn
Ñ Mod
be a Schur module. Define
Φ1 pJ q : XMod Ñ Mod
§7. Homogeneous Functors
by
181
Φ1 pJ qpBX q à
„
|A|n
#A X
J pAq
(where, of course, X is a set, but A ranges over multi-sets). Also, let
α
¸
Ñ BY
syx σyx : BX
P
P
x X
y Y
be a homomorphism. When A
J pAq Ñ J pBq of Φ1 pJ qpαq to be
„ X and B „ Y , we define the component
¸
µ: A
ÑB
sµ J pµq.
This amounts to saying that
Φ1 pJ qpαq ¸
¸
„
„
A X µ: A
B Y
ÑB
sµ J pµq.
Φ1 pJ q is a homogeneous functor of degree n.
Lemma 10.
Proof. Let
α
¸
P
P
szy σzy : BY
Ñ BZ
β
and
y Y
z Z
¸
P
P
tyx σyx : BX
x X
y Y
be homomorphisms, and calculate
rns rn s ¸
¸
b a J s
J
t
cb
ba
b c P
P
b #B
c #C
J
¸
¸ ¸
ÑC
ÑB
ÑC
¸
Ñ
Ñ
A, C B ν : A B
µ: B C
¸¸
A, C B
s µ J
¸
¸
A, B ν : A
sµ tν J pµνq
ν
ÑB
t ν
sµ tν J pµνq
¸
µ: B
µ
¸
A, C B , B 1 ν : A
µ : B1
¸¸
¸
B 1 ,C µ : B 1
P
P
a #A
b #B
ÑC
µ
pq
s J µ
¸
ν: A
ÑB
t J pνq
ν
Ñ BY
182
Chapter 10. Combinatorial Representations
Φ1 pJ qpαq Φ1 pJ qpβq.
At the same time,
rns rns ¸
¸
b a scb
tba
J J c
b
P
P
P
P
rns a #A
b #B
b #B
c #C
¸
a s
t
J cb ba
c a
P
#
A
rns ¸
¸
a J s
t
cb ba
c
P
P
b #B
c #C
P
P
ÑC
¸
A, C ξ : A
b #B
¸
A, C ξ : A
¸
P
a #A
c #C
¸
J
ÑC
¸
P
ξ scb tba ξ
b #B
ξ
¸
P
scb tba J pξq Φ1 pJ qpαβq.
b #B
That Φ1 pJ q is strict polynomial is clear, as the defining equation
¸
Φ1 pJ q P
P
x X
y Y
syx σyx ¸
¸
A, B µ : A
ÑB
sµ J pµq
works when the coeHcients syx belong to any algebra. Finally, it is evident that
it is homogeneous of degree n.
Lemma 11.
ΦpΦ1 pJ qq J.
Proof. The equation
¸
Φ1 pJ q P
P
x X
y Y
syx σyx ¸
„
„
¸
A X µ: A
B Y
ÑB
sµ J pµq
§7. Homogeneous Functors
implies that
Hence
183
Φ1 pJ qσrµs
J pµq.
ΦpΦ1 pJ qqpAq Φ1 pJ q:A pB|#A q Im Φ1 pJ qπrAs
Im J pιA q J pAq,
and
ΦpΦ1 pJ qqpµq Φ1 pJ qσrµs
Lemma 12.
J pµq.
Φ1 pΦpF qq F.
Proof. By the Multi-Cross-Evect Decomposition,
Φ1 pΦpF qqpBX q and
„
|A|n
#A X
à
„
|A|n
#A X
ΦpF qpAq
FA: pB|#A q F pBX q,
¸
Φ1 pΦpF qq à
P
P
x X
y Y
¸
syx σyx ¸
„
„
A X µ: A
B Y
¸
„
„
ÑB
¸
A X µ: A
B Y
sµ ΦpF qpµq
ÑB
sµ Fσrµs
¸
F
P
P
x X
y Y
syx σyx .
Collecting these results together, we obtain the following theorem.
Theorem 9.
The functor
ΦMSetn : Homn
where
Ñ FunpMSetn , Modq,
ΦMSetn pF q : MSetn
takes
rµ : A Ñ Bs ÞÑ
is an equivalence of categories.
Ñ Mod
A ÞÑ FA: pB|#A q
Fσrµs : FA: pB|#A q Ñ FB: pB|#B q ,
184
Chapter 10. Combinatorial Representations
α
–
2ι
α
α
β
σ
β
ι
σ
–
β
ι
σ
–
–
Table 10.2: Multiplication table for MSet2 .
§8. Homogeneous Quadratic Functors
We here determine the structure of Hom2 by classifying the homogeneous
quadratic functors. To find the multi-set description of homogeneous quadratic functors, we first draw the (skeletal) structure of the category MSet2 :
α
&
t1, 1u f
t1, 2u r
σ
β
Every multation reduces to a linear combination of identity multations and
the following:
α
1
1
β
1
2
1
1
σ
2
1
1 2
2 1
The multiplication table is given in Table 10.2. Compare this with Table 10.1
— the only diverence lies in the value of the product βα.
Theorem 10.
A quadratic homogeneous functor is equivalent to the collection of
modules X and Y ; together with homomorphisms α, β, σ as indicated, subject to the
following five relations:
αβ 1
σ,
βσ β,
σα α,
α
Xa
!
Y f
σ2
1,
βα 2.
σ
β
Evidently σ αβ 1 is dispensable. It is enough to have α and β, subject
to the single relation
βα 2.
§9. Schur Modules
The category equivalence
Homn
FunpMSetn , Modq
§9. Schur Modules
185
is yet another instance of Morita equivalence, and we construct a more conceptual proof as follows.
The decomposition
Γn HompBn , q à
„rns
|A|n
ΓA
#A
shows that the functors ΓA constitute a family of small projective generators
for Homn . Letting
L : MSetn Ñ Divn Ñ Homn
denote the contra-variant inclusion functor, there is a Morita equivalence:
p q
Nat L,
*
FunpMSetn , Modq
Homn i
b
L MSetn
As an immediate consequence of this latter view, the functor
Ñ Mod
corresponds to the homogeneous functor F L bMSet J, given by
à
pL bMSet J qpM q LpAqpM q b J pAq
J : MSetn
n
n
P
A MSetn
P
à
A MSetn
ΓA pM q b J pAq.
Imposed upon this module are the following relations:
w b J pµqpxq Γµ pM qpwq b x,
for any x P J pAq, w P ΓB pM q, and multation µ : A Ñ B. Interpreting this
in terms of F, we have the following theorem, where σ denotes the usual
canonical transportation maps.
Theorem 11.
quotient of
Let F be a homogeneous functor of degree n. The module F pM q is the
à
P
A MSetn
by all relations
ΓA pM q b FA: pB|#A q
w b Fσrµs pxq Γµ pM qpwq b x,
for any x P FA: pB|#A q, w P ΓB pM q, and multation µ : A Ñ B.
Moreover, when α : M Ñ N is a homomorphism,
F pαq à
P
A MSetn
ΓA pαq b 1F : pB|
A
#A
q.
186
Chapter 10. Combinatorial Representations
By abuse of notation, we shall often write simply
à
F pM q P
A MSetn
ΓA pM q b FA: pB|#A q,
with the imposed relations tacitly understood. The formula may be extended
to an arbitrary strict analytic functor:
à
F pM q A
À
P
ΓA pM q b FA: pB|#A q.
MSetn
The composition of two strict analytic functors F and G is expressible as
à
F pGpM qq A
À
P
ΓA pGpM qq b FA: pB|#A q,
MSetn
and we obtain the following formula for the multi-deviations of a composition.
Theorem 12.
¸
pF Gqαr s X
A
À
P
¸
P
ω ComA X
MSetn
pGα qbrωs b 1F : pB|
#A
A
q.
Proof. Calculate:
pF Gq
ķ
si b αi
i 1
¸
P
Γ
A
G
A MSetn
¸
A
À
P
MSetn
¸
A
À
P
MSetn
ķ
Γ
A
si b αi
¸
s
B
¸ â
P
B
b 1F : pB|
A
i 1
#A
q
b Gαr s b 1F : pB|
B
A
ä
p qPω
ω a # A a, Z
sZ b GαrZs
(The sum is extended over all ω : A Ñ C, such that #C
Identifying the coeHcient of sX yields the formula.
#A
q
b 1F : pB|
A
#A
q.
„ rks and |C | n.)
Chapter 11
NUMERICAL
STRICT
VERSUS
POLYNOMIAL
FUNCTORS
[. . . ] le plus beau projet de notre académie,
Une entreprise noble et dont je suis ravie,
Un dessein plein de gloire, et qui sera vanté
Chez tous les beaux esprits de la postérité [. . . ].
Molière, Les Femmes savantes
Plainly, the notion of numerical functor is weaker than that of strict polynomial functor. It will naturally be inquired: how much weaker? This chapter
is devoted to a comparison of the two species. The combinatorial and the
module-theoretic viewpoint will be explored in turn, each leading to several
illuminating insights.
The reader should keep in mind that there is a fundamental diverence
between the two types of functors, which is constantly at play. For while
both kinds may be viewed as ordinary functors equipped with extra data, we
know that numerical functors allow for an alternative characterisation, viz. as
ordinary functors satisfying certain equations. A fortiori, a numerical functor
is uniquely determined by its underlying functor. This is not true for strict
polynomial functors, as the following example shows.
Example 1.
Let B be numerical, let A be an algebra (not necessarily numerical!), and let p be a prime. The ring A{pA is a bimodule over B in the usual
way. Keeping the left module structure, equip it with another right module
structure, mediated by the Frobenius map:
px
pAq a ap x
pA.
That this is a module action is a consequence of Fermat’s Little Theorem. Let
pA{pAqp1q denote the bimodule thus obtained.
Define, for any algebra A (not necessarily numerical!), the functors
FA : A XMod Ñ A Mod
M ÞÑ A{pA b M
187
188
Chapter 11. Numerical versus Strict Polynomial Functors
and
GA : A XMod Ñ A Mod
M
ÞÑ pA{pAqp1q b M.
These functors commute with scalar extensions; hence they give strict analytic
functors XMod Ñ Mod.
Let α : M Ñ N be a homomorphism of A-modules, and let a P A. As a
homomorphism
A{pA b M Ñ A{pA b N,
we have
FA paαq 1 b aα a b α aFA pαq,
which shows F is homogeneous of degree 1. As a homomorphism
pA{pAqp1q b M Ñ pA{pAqp1q b N,
we have
GA paαq 1 b aα ap b α ap GA pαq,
which shows G is homogeneous of degree p.
Nonetheless, considered just as (numerical) functors XMod Ñ Mod, F and
G are both linear, and are in fact isomorphic! This is again because of Fermat’s
Little Theorem:
px pBq a ap x pB ax pB,
and consequently
as B-bimodules.
pB{pBqp1q B{pB
4
Let us briefly sum up what is known of numerical versus strict polynomial
functors. For aHne functors, degree 0 and 1, the two notions coincide. This
will no longer be the case in higher degrees, as there exist numerical functors
which do not arise from strict polynomial ones. Even when it exists, the strict
polynomial structure on a given functor is usually not unique. We saw this
in the example above, where it was even possible to define strict polynomial
structures of diverent degrees on the same underlying functor.
The situation for quadratic functors turns out to present an intermediate
case, exhibiting some atypical phenomena. For example, as will be seen below,
the existence of a strict polynomial structure on a quadratic functor may be
inferred from a simple equation.
Another anomaly, occurring in the quadratic case only, is the following.
A quasi-homogeneous quadratic functor may always be made homogeneous of
degree 2; and not only that, but uniquely so. This is singular indeed, and far
from the generic situation. Beware, however, that a quadratic functor which
is not quasi-homogeneous need not arise from a strict polynomial functor.
§1. Quadratic Functors
189
§1. Quadratic Functors
We first propose to examine quadratic functors in detail. Let the following be
the Schur description, as in Theorem 10.10, of a strict polynomial functor F of
degree 2:
α
X1
K
"
X2 c
Y
β
Here K denotes the degree 0 part, X1 the degree 1 part, and βα 2 arises from
degree 2. The labyrinthine description of F is, as may be checked:
α
$
X1 ` Xe 2
K
Y
β
We hence obtain the following characterisation of strict polynomial functors
of degree 2.
Theorem 1.
F:
Let the following be the labyrinthine description of a quadratic functor
α
K
Xa
!
Y
β
• F may be (non-uniquely) extended to a strict quadratic functor iff the following
conditions are satisfied:
– X has a direct sum decomposition X
Im β „ X2 .
X1 ` X2 , such that X1 „ Ker α and
– βα 2X2 .
• F may be (uniquely) extended to a homogeneous quadratic functor iff the following
conditions are satisfied:
0.
βα 2.
– K
–
Example 2.
A continuation of the example above in the case p 2 will
serve to illustrate the theorem, and also to point out its subtlety. Both functors
F and G have the following labyrinthine description:
0
0
B{2B
e
0
"
0
190
Chapter 11. Numerical versus Strict Polynomial Functors
Since it evidently satisfies the conditions of the theorem, it has a unique structure of homogeneous quadratic functor — the functor G above. Yet we managed to exhibit another strict polynomial structure F on this same underlying
functor, but this one linear!
4
It will perhaps be illuminating to write out in detail why it is necessary that
βα 2 in order to define a strict polynomial structure on the functor. This
becomes evident when, given the labyrinthine description of a functor F, we
try to calculate how F acts on the homomorphism
g : B Ñ B,
given by multiplication by the ring element g. The maze associated to g is
G
/ ,
g
and hence

F pgqF : pBqÑF : pBq
¸
„
ΦLaby pF qpGq ΦLaby
P G
g
ΦLaby pF q 1
g 2g βα.
1
/
g
/
g
ΦLaby pF q
2
1
1
/
/
On the other hand, we know what the answer should be for a homogeneous
quadratic functor F, namely
F pgq g2 F p1q g2 .
Equating these two expressions (for g 2) yields βα 2.
Example 3.
Denote by F BrHompB2 , qs2 the fundamental quadratic
functor. There are isomorphisms
F pB2 q BrHompB2 , B2 qs2
BrB22 s2 Brt11 , t12 , t21 , t22 s{J2 ,
where we have denoted
t11
1 0
,
0 0
t12
0
0
1
,
0
t21
0 0
,
1 0
t22
0
0
0
;
1
and J2 is the ideal of polynomials of degree greater than 2. This module is free
of rank 15, and is the direct sum of four cross-evects:
F : pq x1y
2
2
y
F : pB1 q xt11 , t12 , t11
, t11 t12 , t12
§2. The Ariadne Thread
191
2
2
y
F : pB2 q xt21 , t22 , t21
, t21 t22 , t22
F : pB1 |B2 q xt11 t21 , t11 t22 , t12 t21 , t12 t22 y .
It may now be checked that the modules
K
x1y ,
X
xt11 , t12 , t112 , t11 t12 , t122 y ,
Y
xt11 t21 , t11 t22 , t12 t21 , t12 t22 y ,
together with the following maps, constitute the labyrinthine description of
F:
$
$
t11 ÞÑ t11 t21
'
2
'
t11 t21 ÞÑ t11
'
'
'
'
t
Ñ
Þ
t
t
&
&
12
12 22
t11 t22 ÞÑ t11 t12
2
t11
ÞÑ 2t11 t21
α:
β:
t12 t21 ÞÑ t12 t11
'
'
'
'
t11 t12 ÞÑ t11 t22 t12 t21
'
%
2 .
'
t12 t22 ÞÑ t12
%
2
t12 ÞÑ 2t12 t22
We now put
X1
X2
xt112 2t11 , t122 2t12 y
xt112 , t11 t12 , t122 y ,
and observe that, if 2 is invertible in B, X will decomposes as X X1 ` X2 , and
F is in fact strict polynomial of degree 2.
This is an instance of a general phenomenon. We pointed out before
that over a Q-algebra, numerical and strict polynomial functors coincide.
A slightly stronger statement is true: If the integers 1 through n are invertible,
then numerical and strict polynomial functors of degree n coincide. This may not be
completely obvious from the theory developed thus far, but we shall prove it
presently.
4
§2. The Ariadne Thread
Before Theseus entered the legendary labyrinth to fight the Minotaur, Ariadne
presented him with a wonderful gift: the thread that will now forever bear
her name. This device would eventually assist him in backtracking out of the
frightful maze. Such is the legend. We too will be assisted on our quest by an
Ariadne thread.
Let us expound our doctrine. We know that homogeneous functors of
degree n correspond to Schur modules
J : MSetn
Ñ Mod,
and numerical functors to labyrinth modules
H : Labyn
Ñ Mod.
Pre-composition with the Ariadne functor corresponds in evect to the forgetful functor
Homn Ñ Numn .
192
Chapter 11. Numerical versus Strict Polynomial Functors
In fact, since homogeneous functors are of course quasi-homogeneous, we
have the following diagram:
D
jj5 Mod
jjjj j
jj X
jjjj
j
j
j
j
Forget
Hom
LabynI
n
II
II
II
II
An
II
$
MSetn
j
QHomj
n jj
Theorem 2: The Ariadne Thread.
1
ΦLaby Φ
MSetn
pAn q .
Proof. We must show that, for a Schur module J : MSetn
Ñ Mod,
1
ΦLaby Φ
MSetn pJ q JAn .
Denoting H
1
ΦLaby ΦMSet
pJ q, we have, for a finite set X,
n
H pX q Φ1
MSetn
Im
¸
pJ q: pB|X q Im Φ1
MSetn
¸
1
p1q|X ||Y | ΦMSet
pJ q n
„
Y X
Recalling that
pJ q
1 p J q
ΦMSet
n
P
♦ πx
P
x X
πy .
y Y
¸
syx σyx
¸
¸
A, B µ : A
ÑB
sµ J pµq,
we obtain
H pX q Im
¸
„
p1q|X ||Y |
Y X
Im
¸
„
„
|A|n
J pιA q
#A Y
p1q|X ||Y |
Y X
Im
¸
¸
|A|n
#A X
¸
„
|A|n
#A Y
1J pAq
à
|A|n
#A X
1J pAq
J pAq JAn pX q.
The fourth step is due to the Principle of Inclusion and Exclusion.
§3. Numerical versus Strict Polynomial Functors
193
We now turn to computing H pP q when P : X Ñ Y is a maze. It will be
enough to consider the case of a simple maze, as any maze may be expressed
as a sum of such. Denote the passages of P by pi : xi Ñ yi for 1 ¤ i ¤ k, and
compute:
1
H pP q Φ
MSetn pJ q ♦ pi σyi xi
¸
I
1
p1q |I | ΦMSet
pJ q
n
k
„rks
¸
pi σyi xi
,
p1qk|I |
¹
P
i I
of which the J pAq Ñ J pBq component is
¸
I
„rks
p1qk|I |
¹
¸
µ: A
ÑB
pIµpaqa J pµq
where we have defined
pIba
#
¸
µ: A
ÑB
¸
I
„rks
pIµpaqa J pµq, (1)
if a xi and b yi for i P I
else.
pi
0
We see that for the coeHcient of J pµq to be non-zero, all elements of the multation µ must correspond to passages in P. The converse also holds, namely,
that all passages of P must be represented in µ. This is because of the following
reason. If a passage pj be “missing” from µ, sets I with and without j in (1) will
give rise to terms of alternating signs, which will cancel. Hence the coeHcient
of J pµq will survive only if µ is of the form
µ
¹ x r mi s
i
i
for positive integers
m1
Then only I
yi
mk
,
n.
rks will yield a non-zero contribution in (1), and consequently
H pP q ¹
mi
¸
m1
mk n
pi
J
¹ x r mi s
i
i
yi
JAn pPq.
§3. Numerical versus Strict Polynomial Functors
As the crowning glory of our work, the pinnacle of the palace, let us record
the exact obstruction for a numerical functor to be strict polynomial. Recall
194
Chapter 11. Numerical versus Strict Polynomial Functors
from Theorem 9.4 the homomorphism
εn : Γn pBnn q Ñ Q bZ BrBnn sn
σrAs ÞÑ
1
♦σ
deg A A
and the direct sum decomposition
Im εn
Γn pM q ` pKer γn X Im εn q.
Theorem 3: The Polynomial Functor Theorem.
Let F be a quasi-homogeneous functor of degree n, corresponding to the labyrinth module
H : Labyn
Ñ Mod
and the BrBnn sn -module M. The following three constructions are equivalent:
A. Imposing the structure of homogeneous functor upon F.
B. Exhibiting a factorisation of H through Laby`n .
C. Giving M the structure of Im εn -module.
Proof. The isomorphism Laby`n
From the isomorphism
Im εn
MSetn shows the equivalence of A and B.
Γn pBnn q pKer γn X Im εn q
we conclude that Γn pBnn q-modules canonically correspond to Im εn -modules,
and vice versa. (The ring Ker γn X Im εn corresponds to subfunctors of lower
degree. By considering quasi-homogeneous functors only, modules over this
ring will be zero.) This shows the equivalence of A and C.
We caution the reader that, even in the case H factors through Laby`n and
M may be considered an Im εn -module, the factorisation and the module structure are not unique. There are in general many strict polynomial structures
on the same functor, even of diverent degrees!
Example 4.
We point out one particular case when any such H will factor
(uniquely) through Laby`n . When B is a Q-algebra,
Laby`n
MSetn
is simply the additive hull of Labyn . This mirrors the already well-known fact
that, over a Q-algebra, numerical and strict polynomial functors coincide. In
fact, as is seen from the definition of Laby`n , it is suHcient for the integers 1
through n to be invertible in B to guarantee such a factorisation.
4
§4. Restriction and Extension of Scalars
195
Example 5.
For affine functors (degree 0 and 1), numerical and strict polynomial functors coincide. This is no longer the case in higher degrees. Yet, as
we saw before, the quadratic case retains some regularity, in that any quasihomogeneous functor is necessarily homogeneous. Any quasi-homogeneous
functor of degree 2 may be given a unique strict polynomial structure, which
makes it homogeneous of degree 2.
The reason for this becomes clear when we examine the creation of Laby`2
in detail. The localisation procedure requires us to adjoin mazes
1
2
/
1
/
1
,
but in Laby2 , the equation
1
/
1
/
2 1
/
holds, as may be verified. Hence Laby`2 is simply the additive hull of Laby2 . 4
§4. Restriction and Extension of Scalars
The divided power map
γn : BrBnn sn
Ñ Γn pBnn q
gives rise to two natural functors between the corresponding module categories, namely restriction and extension of scalars. We consider them in turn.
Restriction of scalars is the functor
Γn Bnn
p
q Mod Ñ BrBnn sn Mod,
which takes a Γn pBnn q-module M and views it a BrBnn sn -module under the
multiplication
rσsx γn pσqx σrns x F pσqx.
On the functorial level, this corresponds to the forgetful functor
Homn
Ñ Numn .
Extension of scalars is the functor
B Bnn n Mod
r
s
Ñ Γ pB q Mod,
n
n n
which takes a BrBnn sn -module M and transforms it into a Γn pBnn q-module
Γn pBnn q bBrBnn sn M
through the tensor product. How does it act on the functorial level?
196
Chapter 11. Numerical versus Strict Polynomial Functors
Let us denote by
P BrHompBn , qsn
Q Γn HompBn , q
the projective generators of the categories Numn and Homn , respectively. A
functor F P Numn will then correspond to the BrBnn sn -module
NatpP, F q.
Extension of scalars transforms it into the Γn pBnn q-module
N Γn pBnn q bBrB s M Γn pBnn q bBrB s NatpP, F q,
M
n n n
n n n
which corresponds to the homogeneous functor
G Q bΓn pBnn q N
Q bΓ pB q Γn pBnn q bBrB s
Q bBrB s NatpP, F q.
n
n n
n n n
NatpP, F q
n n n
This tensor product is interpreted in the usual way. By definition,
P
Þ Q bBrB s NatpP, Pq
Ñ
Q bBrB s BrBnn sn Q,
n n
n n
n
n
and we then extend by direct sums and right-exactness.
We summarise in a theorem.
Consider the divided power map
Theorem 4.
γn : BrBnn sn
Ñ Γn pBnn q.
• Restriction of scalars
Γn Bnn
p
q Mod Ñ BrBnn sn Mod
corresponds to the forgetful functor
Ñ Numn .
Homn
• Extension of scalars corresponds to the functor
B Bnn n Mod
r
s
Ñ Γ pB q Mod,
n
n n
which maps
F
ÞÑ Γn HompBn , q bBrB s
n n n
NatpBrHompBn , qsn , F q.
§5. Torsion-Free Functors
197
Restriction of scalars, as we know, corresponds on the combinatorial level
to the Ariadne functor. Extension of scalars, on the contrary, does not seem
to admit a simple combinatorial interpretation. Evidently, this procedure
will grasp any numerical functor, and through brute force transform it into
a homogeneous functor of degree n. The functor to start with need not be
quasi-homogeneous; it may very well contain parts of lower degree. Excessive
amounts of violence will then be needed to transfigure it into a homogeneous
functor, and it is hardly surprising that this process will wreak havoc with its
internal structure.
§5. Torsion-Free Functors
In this section we consider torsion-free functors and modules. Torsion shall, as
before, always mean Z-torsion. We let
M
denote the greatest torsion-free quotient of the module M, which is simply M
divided by its torsion submodule. Observe that
Q bZ M Q bZ M.
Also, when C is a category of modules or functors, we shall let
C
denote the subcategory of torsion-free modules.
Definition 1.
Let A, B „ C be linear categories. The category of (epic)
angles joining A and B over C is the linear category of formal angles
X
/Zo
Y ,
X
P A, Y P B, Z P C;
with both arrows epic (in C). It will be denoted by
A _C B.
We shall usually suppress mention of the epics, and denote the formal
angle
/Zo
X
Y
by simply
rXZY s,
when no confusion is likely to result.
198
Chapter 11. Numerical versus Strict Polynomial Functors
An arrow in the angle category is defined in the obvious way. Namely, an
arrow from rXZY s to rX 1 Z1 Y 1 s is given by a commutative diagram:
X
/Zo
Y
X1
/ Z1 o
Y1
where the three arrows X Ñ X 1 , Y Ñ Y 1 , and Z Ñ Z1 belong to A, B, and C,
respectively.
A subtle fact, which has hitherto gone unmentioned, is the following. A
careful examination of the proof of Theorem 9.7 will reveal that
BrHompBn 1 , qsn
is another projective generator for Numn . As a notable consequence, the rings
BrBnn sn1
and BrBpn1qpn1q sn1
are Morita equivalent, so that, in fact,
Numn1
BrBp qp q s Mod BrB s Mod.
n 1
n 1
n n
n 1
n 1
The set-up is then as follows. We have rings
R BrBnn sn
BrBnn sn1
T Im γn „ Γn pBnn q,
S
and surjections
σ : R Ñ S,
τ: R Ñ T,
where σ denotes the canonical quotient map, and τ γn . By Theorem 9.2, the
homomorphism
pσ, τq : R Ñ S T
is an injection of finite index. The goal is to show how this leads to a category
equivalence
R Mod S Mod _R Mod T Mod .
(Recall the construction of the pullback categpry in Chapter 0.)
Let us begin with a very general remark. Let χ : A Ñ B be a surjective ring
homomorphism, and let M be an A-module. There is an exact sequence:
0
The module
/ pKer χqM
B bA M
/M
/ B bA M
/0
M {pKer χqM
is the greatest B-quotient of M. Furthermore, pB bA M q is the greatest torsionfree B-quotient. Applying this observation to the homomorphisms σ : R Ñ S
and τ : R Ñ T , we conclude:
§5. Torsion-Free Functors
Theorem 5.
QHomn .
Theorem 6.
199
Every functor in Numn has a greatest quotient in both Numn1 and
When rXZY s is an object of the category
_
S Mod
R Mod
,
T Mod
the module Z is a torsion module.
Proof. There are R-homomorphisms:
X
ϕ
ψ
/Zo
Y
Note that X and Y are also S T -modules, according to:
ps, tq x sx,
ps, tqy ty.
Consider a z P Z. Because R has finite index in S T , we can find an integer
p 0 such that pp1, 0q P R. Choose x P X such that
ϕpxq z,
and calculate
pz pϕpxq ϕppxq
ϕppp, 0q xq pp, 0qϕpxq pp, 0qz.
In the fifth step, we used that ϕ is R-linear. Similarly, there is a non-zero
integer q such that p0, qq P R and
qz p0, qqz.
We conclude that
pqz pp, 0qp0, qq z 0 z 0,
so that z is a torsion element.
We shall now construct our category equivalence:
R Mod
j
Π
*
_
S Mod
R Mod
T Mod
Σ
Consider first an M
P R Mod . There is a homomorphism of R-modules
χ : M Ñ pS bR M q ` pT bR M q
x ÞÑ p1 b x, 1 b xq.
200
Chapter 11. Numerical versus Strict Polynomial Functors
The projections
M Ñ S bR M
x ÞÑ 1 b x
M
Ñ T bR M
x ÞÑ 1 b x
are onto, because σ and τ are onto.
Let us now show χ is one-to-one. Consider the following exact sequence
of R-modules:
/ p S bR M q ` p T
χ
/M
/ Ker χ
0
bR M q Tensoring with the Z-module Q produces:
0
/ Ker χ bZ Q
/ M bZ Q
b /
p S bR M q ` p T bR M q bZ Q
χ 1
/ pS T q bR M bZ Q
R bR M bZ Q
Since R has finite index in S T , the lower map is an isomorphism, and hence
Ker χ bZ Q 0. We conclude that Ker χ is a torsion module, and therefore,
being included in the torsion-free module M, zero. Consequently, χ is one-toone.
We thus infer, by Delsarte’s Lemma, the existence of a formal angle
p S bR M q ϕ
/
pS bR M q ` pT bR M q {M o
in the category
We define this to be ΠpM q.
_
S Mod
R Mod
Π: M
pT bR M q
.
T Mod
The functors
Theorem 7: The Splitting-Off Theorem.
ψ
ÞÑ pS bR M q , pS bR M q ` pT bR M q {M, pT bR M q
Σ : rXZY s ÞÑ KerpX ` Y Ñ Zq
provide a category equivalence:
R Mod
j
Π
*
_
S Mod
R Mod
Σ
Consequently, there is an equivalence of functor categories:
Numn
QHomn _Num
n
T Mod
Numn1 .
§5. Torsion-Free Functors
201
Proof. With notation as above,
Kerpϕ
ψq M,
so we immediately have ΣΠ I.
Consider now an angle
rXZY s P S Mod _ Mod T Mod .
R
Defining
ΣprXZY sq KerpX ` Y Ñ Zq,
we observe that M has finite index in X ` Y , since Z is torsion. Furthermore,
M
ΠΣprXZY sq ΠpM q pS bR M q , pS bR M q ` pT bR M q {M, pT bR M q
.
By Delsarte’s Lemma, X is an S-quotient of M, and, moreover, it is by
assumption torsion-free. Since pS bR M q is by definition the greatest torsionfree S-quotient of M, there is a factorisation:
0
/ Ker ξ
M II
II
II
II
II
I$
/X
/ pS bR M q
ξ
/0
Tensoring with Q yields:
0
/ Ker ξ bZ Q
M bZ Q
OOO
OOO
OOO
OOO
'
/ p S bR M q bZ Q
/ X bZ Q
b
ξ 1
/0
Because M has finite index in X ` Y , we have
M bZ Q pX bZ Qq ` pY
bZ Qq.
We infer that the greatest torsion-free S-quotient of M bZ Q is, in fact, X bZ Q.
Therefore, the homomorphism ξ b 1 is an isomorphism, with kernel
Ker ξ bZ Q 0.
The module Ker ξ is torsion, and therefore zero, being included in the torsionfree module pS bR M q . We have thus deduced
pS bR M q X.
Similarly,
pT bR M q Y ,
202
Chapter 11. Numerical versus Strict Polynomial Functors
and there only remains the verification
pS bR M q ` pT bR M q {M pX ` Y q{ KerpX ` Y Ñ Zq
ImpX ` Y Ñ Zq Z.
The relation ΠΣ I follows.
Chapter 12
POLYNOMIAL
MONADS
AND
OPERADS
Han ämnar nu ge ut, till vinst för Pappersbruken,
Till vinst för hvar och en som läser i vårt land,
Et dubbelt Skalde verk ifrån sin lärda hand:
Om rynkbands nytta i peruken
Och skadan utaf mal i gamla folioband.
Anna Maria Lenngren, Herr Grälberg
This closing chapter will deal with practical matters — it is the palace kitchen,
if the expression be us pardoned. Let us, very briefly, indicate how polynomial
functors could possibly be put to use.
The theory of operads has long suvered from its unnatural restriction to
fields of characteristic 0. Polynomial functors, it turns out, allow for a natural
extension to arbitrary base rings. We certainly do not aim to be encyclopædic,
but merely to sketch an outline of what polynomial monads and operads look
like, and how they behave under the fundamental operations of induction
product and plethysm.
It is our hope that somebody, someday, might find this theory useful.
§1. Classical Operads
The reader wishing a comfortable introduction to operads is referred to the
manuscript [15] by Professors Loday and Vallette (currently at the draft stage),
which has served as our source of information.
Let B be a field of characteristic 0, and let Σn denote the symmetric group
on n symbols. A Σn -module is (by definition, if you like) the same as a functor
BrΣn s Ñ Mod,
where BrΣn s is interpreted as a category with a single object. Already in 1995,
Professor Macdonald had established a category equivalence
Homn
Σ Mod.
n
203
204
Chapter 12. Polynomial Monads and Operads
A homogeneous functor F would correspond to the Σn -module1
NatpT n , F q,
and a Σn -module P to the homogeneous functor
Þ T n pM q bΣ P.
Ñ
on T n pM q is the obvious one.)
M
n
(The right action of Σn
See [16], in which
the result has been relegated to an “appendix”! This led to the following
definitions.
A Σ-module is a family
Definition 1.
P
pPn qnPN
of Σn -modules.
The Schur functor associated to a Σ-module P is
Definition 2.
XMod Ñ Mod
M
ÞÑ
8
à
n 0
T n pM q bΣn Pn .
A classical operad is a Σ-module with an associated monadic
Definition 3.
Schur functor.
Theorem 1.
Over a field of characteristic 0, a classical operad is equivalent to a
strict analytic monad.
Proof. Because of the category equivalence
Homn
Σ Mod,
n
giving a strict analytic monad is the same as specifying a Σ-module, and requiring that its Schur functor be monadic.
Operads and monads are thus equivalent, and we shall usually prefer the
latter viewpoint. The examples we give below will all be of monads.
When considering a strict analytic monad
D : XMod Ñ Mod,
we have automatically the concept of algebra (as inherited from the theory of
monads). The base category is (as always) Mod, so an algebra for our operads
will first of all be a module.
p
q
1 Evidently, Nat T n , F is a right module over Nat T n . That indeed Nat T n
that dates back to Weyl.
BrΣn s is a result
§1. Classical Operads
Example 1.
ear functor
205
Let A be an (associative, unital) algebra, and consider the linDpM q A b M.
It is canonically a monad, with structure maps
µA : A b A b M Ñ A b M,
εA : M Ñ A b M,
a b b b x ÞÑ ab b x
x ÞÑ 1 b x.
An algebra over this monad is a homomorphism
AbM
ÑM
(satisfying some axioms), which is just an A-module. D is the operad (or monad) of A-modules.
4
Example 2.
The tensor functor T is a strict analytic monad. An algebra
over T is a homomorphism
T pM q Ñ M,
which is just an associative, unital algebra. T gives the operad of associative
algebras. (To obtain non-unital algebras, simply remove the degree 0 part
from T , which produces the reduced tensor algebra.)
4
Example 3.
The symmetric functor S is a strict analytic monad. An algebra over S is an associative, commutative, and unital algebra. The operad
corresponding to S is known as the operad of commutative algebras.
4
Example 4.
gebras.
The monad Λ is evidently the operad of anti-commutative al4
Example 5.
The monad Γ is the operad of divided power algebras. In the
current context, B being a field of characteristic 0, Γ S.
4
Example 6.
The functor L, which constructs the free Lie algebra on a module, is a monad. Being a subfunctor of T , it is strict analytic. It gives the operad
of Lie algebras.
4
There are two standard constructions on operads: the induction product
and the plethysm. On the monadic level, they are simply tensor product and
composition, respectively.
Definition 4.
Let P and Q be operads with Schur functors F and G, respectively. The induction product is the Σ-module
PbQ
that has Schur functor F
b G.
206
Chapter 12. Polynomial Monads and Operads
There is a conceivable risk of confusion here with the usual use of the
symbol b, but we shall never have occasion to consider the tensor product of
Σ-modules.
It is contended in [15] that the induction product is given by the formula
P b Q: X
ÞÑ
à
\
A B X
P pAq b QpBq.
We shall presently produce a stronger statement, from which this formula can
be derived as a special case.
Definition 5.
Let P and Q be operads with Schur functors F and G, respectively. The plethysm is the Σ-module
PQ
that has Schur functor F
G.
To describe the plethysm, we recall the terminology of compositions from
Chapter 2. Let X be a set, and let ω : rns Ñ 2X be an rns-composition of X.
Define
Qpωq Qpωp1qq b Qpωp2qq b b Qpωpnqq.
According to [15], the plethysm is given by the formula
P Q: X
ÞÑ
8
à
P
à
n 0 ω Comrns X
P prnsq b Qpωq.
§2. Schur Operads
Arguably, the fields of characteristic 0 form a rather limited class of rings. It
should be clear, then, that an extension of the operad concept is called for.
Evidently, there is no problem in considering strict analytic monads over an
arbitrary (commutative and unital) base ring, but it is, perhaps, not as evident
what the appropriate generalisation on the operad side is.
A Σ-module is (by definition, if you like) the same as a functor
BrΣn s Ñ Mod,
where BrΣn s is interpreted as a category with a single object. It extends to a
linear functor
Setn Ñ Mod,
where Setn denotes the linear category of sets of cardinality n. It was the insight of Dr. Salomonsson and Professor Ekedahl ([20]) that, over an arbitrary
commutative and unital base ring, Setn is superseded by MSetn . These two
categories are, quite blatantly, not equivalent, but they are Morita equivalent
over a field of characteristic 0.
§2. Schur Operads
Definition 6.
207
A Schur module is a linear functor
8
à
MSetn
n 0
Ñ Mod.
Recall the category equivalence
ΦMSetn : Homn
Ñ FunpMSetn , Modq,
under which a homogeneous functor F corresponds to the Schur module
ΦMSetn pF q : X
ÞÑ NatpΓX , F q.
Definition 7.
The Schur functor associated to a Schur module P is the
homogeneous functor
1
Φ
MSetn pP q : XMod Ñ Mod
M
à
ÞÑ
X
À
P
ΓX pM q b P pX q.
MSetn
Definition 8.
A Schur operad is a Schur module with an associated monadic Schur functor.
We then have the following result, which appears as Theorem I.3.4 in
Dr. Salomonsson’s thesis [20].
Theorem 2.
Over a commutative, unital base ring, a Schur operad is equivalent to
a strict analytic monad.
Proof. Immediate from the category equivalence above.
Example 7.
Let B be a ring of characteristic p. Recall from Professor Jacobson’s excellent treatise on Lie algebras, [13], the definition of a restricted Lie
algebra of characteristic p. It is a Lie algebra A of characteristic p, equipped with
a unary operation pqrps , satisfying the following axioms:
1. paxqrps
2. px
ap xrps , for a P B, x P A.
yqrps
xr p s
y rp s
p¸1
si px, yq, where isi px, yq is the coeHcient of ti1
i 1
in the expansion of xpadptx yqqp1 . Here x and y commute with the
indeterminate t, but not with each other.
3. rxyrps s xpad yqp , for x, y P A.
208
Chapter 12. Polynomial Monads and Operads
The monad Lp of free restricted algebras is strict analytic, and gives rise to the
operad of restricted Lie algebras of characteristic p. It is not given by a classical
operad, mainly because the characteristic is wrong.
4
Example 8.
No treatment of operads with self-respect could omit mention of trees. Like their classical counterparts, also Schur operads may be
illustrated by trees. An element
pur2s v b vwq b x P Γt1,1,1,2,2u pM q b Ppt1, 1, 1, 2, 2uq
of the Schur functor should be thought of as a multi-tree2 :
xIT
jujujuju ITITITITTTT
j
j
j
1
1jjjj 1 uu
I 2 TTT2
u
u
v
v
w
From their interpretation as a tensor product, it is clear that multi-trees
are linear in each vertex (top and bottom row, that is; those in the middle
are just labels). But, as we recall from Theorem 10.11, they are subject to yet
another relation, which we now exemplify. Let u, v P M, and let x P P pt1, 1, 2uq.
Because
1 2 2 1 1 2
1 1 2
1 1 2
2
,
u u v 1 2 2
u u v
u v u
we have a relation
pu b uvq b P
1
1
1 2
2 2
on the Schur functor
pxq 2pur2s b vq b x puv b uq b x
à
X
À
P
ΓX pM q b P pX q,
MSetn
and a corresponding equation for trees, which might be termed “vertical associativity”:
1 1 2
x
1 u2uIII2
I
u
II 2
1 uuu 2
u
u
v
2 x
uII
uu 1 III
u
1 uu
I2
u
u
v
x
uII
uu 2 III
u
1 uu
I1
u
u
v
4
2 This was originally Dr. Salomonsson’s device, and somewhat (genetically) modified to suit
our needs.
§3. The Induction Product of Schur Operads
209
§3. The Induction Product of Schur Operads
Induction products and plethysms of Schur modules are defined as for Σ-modules. We emphasise again that these operations correspond to tensor product
and composition of the corresponding Schur functors.
Consider two Schur modules P and Q, corresponding to Schur functors F
and G, respectively. Let π, as usual, denote projection. According to Theorem
7.8, the induction product P b Q maps
pP b QqpX q pF b Gq:X pB|#X q
ImpF b Gqπr s X
à
\
A B X
F : pB|
#X
A
à
Im FπrAs
\
q b G: pB|
A B X
B
#X
q
b Gπr s
B
à
\
A B X
P pAq b QpBq.
The action of a multation µ : X Ñ Y is found as follows. By the Multi-Set
Yoneda Lemma, it corresponds to
pF b Gqσr s : pF b Gq:X pB|#X q Ñ pF b Gq:Y pB|#Y q,
µ
or, as it so happens, the homomorphism
pF b Gqσr s :
µ
à
\
A B X
FA: pB|#X q b G:B pB|#X q Ñ
à
\ C D Y
FC: pB|#Y q b G:D pB|#Y q,
given by the formula
pF b Gqσr s µ
Theorem 3.
¸
\
κ λ µ
Fσrκs
b Gσr s .
λ
The induction product of two Schur modules P and Q is given by
pP b QqpX q For a multation µ : X
Ñ Y,
pP b Qqpµq à
\
A B X
¸
\
κ λ µ
P pAq b QpBq.
P pκq b Qpλq.
§4. The Plethysm of Schur Operads
We now turn to the plethysm. From Theorem 10.12, we recall the formula
pF Gqαr s ¸
X
A
¸
P
â
P
ä
p qPω
ω ComA X a #A a,Z
GαrZs b 1F : pB|
A
#A
q.
210
Chapter 12. Polynomial Monads and Operads
This yields the plethysm
pP QqpX q ImpF Gqπr s
Im
¸
A
Im
¸
¸
â
P
P
¸
â
P
à
ä
P
p qPω
#A
A
q
ä
P
ω ComA X a,Z
A
GπrZs b 1F : pB|
p qPω
πG: pB|
p qPω
ω ComA X a #A a,Z
ä
ω ComA X a #A a,Z
A
à
X
#Z
Z
QpZq
q
b 1F : pB|
#A
A
q
b PpAq,
with certain relations divided away, as seen in the following theorem.
Theorem 4.
The plethysm of two Schur modules P and Q is given by
pP QqpX q à
à
ä
P
P
p qPω
ω ComA X a,Z
A MSet
QpZq
b PpAq,
which is a quotient by all relations
ä
pb,ZqPω
wZ
b Ppνqpxq ä
pa,ZqPων
wZ
b x,
for any elements x P P pAq, wZ P QpZq, B-composition ω, and multation ν : A Ñ B.
For a multation µ : X Ñ Y ,
pP Qqpµq ¸
¸
ä
p qP
P
ξ ComA µ a,ν ξ
A
Qpνqb 1PpAq .
Proof. Only the last part remains:
pP Qqpµq pF Gqσr s
¸
A
¸
A
µ
¸
â
P
P
ä
p qP
ξ ComA µ a #A a,ν ξ
¸
P
ä
p qP
ξ ComA µ a,ν ξ
Gσrνs b 1F : pB|
A
Qpνqb 1PpAq .
#A
q
§5. Sur Operads
211
As an example of a relation of the above kind, we take
Example 9.
X t1, 1, 2u
A ta, a, bu
B tc, d, d u,
and
ν
Because
a a
c d
a
ων 2
t1u
ω
b
,
d
a
t1 u
b
t2u
c
t1u
a
t1u
d
t1u
d
t2u .
a
t2u
b
t1 u ,
we get
pw1 b w1 w2 q b Ppνqpxq 2pwr12s b w2 q b x pw1 w2 b w1 q b x,
where the left-hand side comes from
ä
ps,ZqP t u t u t u
c
1
d
1
d
2
QpZq b P pBq Qpt1uq b Qpt1uq b Qpt2uq b P pBq,
and the right-hand side from
ä
ps,ZqP ta1u ta1u tb2u
QpZq b P pAq ` Γ2 pQpt1uqq b Qpt2uq b P pAq
`
ä
ps,ZqP ta1u ta2u tb1u
QpZq b P pAq
Qpt1uq b Qpt2uq b Qpt1uq b P pAq .
4
§5. Sur Operads
The theory of numerical functors and mazes allows for a further extension of
the operad concept.
Definition 9.
A labyrinth module is a linear functor
Laby Ñ Mod.
Recall the category equivalence
ΦLaby : FunpXMod, Modq Ñ FunpLaby, Modq,
under which a module functor F corresponds to the labyrinth module
ΦLaby pF q : X
ÞÑ Natp∆X , F q.
212
Chapter 12. Polynomial Monads and Operads
Definition 10.
module functor
The Sur functor associated to a labyrinth module P is the
1
Φ
Laby pP q : XMod Ñ Mod
M
ÞÑ
à
P
X Laby
∆X pM q b P pX q.
Definition 11.
A Sur operad is a labyrinth module with an associated
monadic Sur functor.
Quite analogous to the situation for Schur operads, we have the following
result.
Theorem 5.
Over any (unital) base ring, a Sur operad is equivalent to a monad.
Proof. Immediate from the category equivalence above.
For our purposes, it will not be necessary to restrict attention to analytic
monads, but it might well be needed for any sensible applications.
Example 10.
Sur operads manage to capture at least two additional classes
of algebras. First, of course, there is the monad
N pM q B
M
of free numerical rings. It gives the operad of numerical algebras.
4
Example 11.
As another example, Sur operads succeed in encoding λrings. Constructing the free λ-ring on a module yields an analytic monad,
which we call the operad of λ-algebras.
4
Example 12.
element
Let us now describe the graphical version of Sur operads. An
1
vv zvvu
pdH v b x P ∆t1,2u pM q b P pt1, 2uq
U H
vHH
2
w
u
of the Sur functor should be thought of as a qvast3 :
x
uII
uu III
u
I
uu
1I
2I
II
uu III
u
II v v III w
u uuu u
3 The
Swedish word qvast means broom-stick.
§6. The Induction Product of Sur Operads
213
The qvast is, unlike the multi-tree, not linear in its vertices or edges. Rather,
being a maze construct itself, it will inherit its properties from the labyrinth
category, which we leave for the interested reader to find.
Let us, however, indicate an instance of the relation displayed in Theorem
10.7. Let u, v P M, a right module over the possibly non-commutative ring B,
and let x P P pt1, 2uq. Because
v1
oa
s s
vav
v
z
v
ysysssv 1 b vv 1 ,
pdH
v
eKK
ub
v
{
H
KK
HH 2 oc 2
u
uc
2
2
u ss
1
ua
there will be a relation
1
a
vv s s
1 o v1
zvvva
ysysssv p
b P b{vvv pxq eKK
dHHH ub b x
KK
2 oc 2
H
u
uc
2
2
u ss
on the Sur functor
1
à
P
X Laby
ua
∆X pM q b P pX q,
which will again translate into vertical associativity:
xB
|| BBB
|
B
||
2
1P
PPP
PP
a
bPP c
1
2
|BBB
|
BB v u
u |||
xB
|| BBB
|
B
||
1
2
|BBB
|
uc
ua ||| va BB ub
.
4
§6. The Induction Product of Sur Operads
Induction products and plethysms of labyrinth modules are defined as in the
classical case.
Let us find the induction product. Consider two labyrinth modules P and
Q, corresponding to Sur functors F and G, respectively. According to Theorem 7.2, the induction product P b Q maps
pP b QqpX q pF b Gq: pB|X q
214
Chapter 12. Polynomial Monads and Operads
ImpF b Gq
♦ πx
P
x X
à
♦ πa
Im F
bG
♦ πb
aPA
bPB
à
:
:
F pB|A q b G pB|B q P pAq b QpBq.
AY B X
AY B X
Yà
A B X
The induction product of two Sur modules P and Q is given by
Theorem 6.
pP b QqpX q For a maze R : X
à
Y
A B X
Ñ Y,
¸
pP b QqpRq P pAq b QpBq.
P pSq b QpT q.
Y S T R
§7. The Plethysm of Sur Operads
One last definition, before we reach the end. When X is a set and pMy qyPY is a
(finite) family of modules, we shall let
∆
X
ppMy qyPY q „ ∆
X
à
P
My
y Y
denote the module of deviations r♦k uk s such that every uk belongs to some My ,
each My being represented at least once. Alternatively, this may be interpreted as
the set of mazes X Ñ , where the passages have been labelled with elements
of the modules My , again with each module represented at least once.
Lemma 1.
Let
à
πy :
P
My
y Y
à
Ñ
P
My
y Y
be the canonical projections, and consider
¤
P
"
πy
/
*
Im ¤
P
y Y
"
:∆
X
πy
à
P
y Y
Its image is
y Y
*
My
Ñ∆
X
à
P
y Y
/ ∆X ppMy qyPY q.
My
.
§7. The Plethysm of Sur Operads
215
From Theorem 10.8 we have the formula
¸
pF Gqpα1 αn q ¸
P
rs
A Laby J C n
"
¤
P
p
G ♦iPI αi
q /
*
I J
b 1F : pB|
A
q,
which we use to compute the plethysm:
pP QqpX q ImpF Gq
♦ πx
P
x X
¸
Im
¸
P
A Laby W CX
¸
Im
P
¸
à
P
A Laby W CX
à
à
P
A Laby W CX
P
Z W
¤
#
p
G ♦zPZ πz
π
G
q /
: pB|Z q
/
+
+
P
A
:
:
∆ pG pB|Z qqZPW b F pB|A q
A Laby W CX
à
#
¤
Z W
b 1F : pB|
q
b 1F : pB|
q
A
A
∆A ppQpZqqZPW q b P pAq,
with certain relations divided away, as displayed in the following theorem.
Theorem 7.
The plethysm of two Sur modules P and Q is given by
pP QqpX q à
à
P
A Laby W CX
∆A ppQpZqqZPW q b P pAq,
which is a quotient by all relations
b PpRqpxq UR b x,
for any W C X, elements x P P pAq, U P ∆B ppQpZqqZPW q, and maze R : A Ñ B.
For a maze S : X Ñ Y ,
U
¸
pP QqpSq ¸
P
A Laby ECS
¤
"
P
p q/
Q T
*
T E
b 1PpAq .
Proof. There remains only the last statement:
pP QqpSq pF Gq
¸
¸
P
A Laby ECS
¸
P
♦
rs : xÑysPS
¸
A Laby ECS
¤
P
T E
¤
P
T E
#
"
sσyx
p
G ♦rs : xÑysPT sσyx
Q T
p q/
*
q/
+
b 1PpAq .
b 1F : pB|
A
q
216
Chapter 12. Polynomial Monads and Operads
Example 13.
From the classical case of operads and trees, the graphical interpretation of the induction product and the plethysm is clear. The induction
product corresponds to a two-stemmed tree, and the plethysm to a grafted
tree. These interpretations remain valid for multi-trees and qvastar.
4
Bibliography
[1] Baues, Dreckmann, Franjou & Pirashvili: Foncteurs polynômiaux et foncteurs
de Mackey non linéaires, Bulletin de la Société mathématique de France
no 129, fascicule 2, 2001.
[2] Francis Borceux: A Survey of Semi-Abelian Categories, Fields Institute Communications, Volume 43, 2004.
[3] Serge Bouc: Non-Additive Exact Functors and Tensor Induction for Mackey Functors, Memoirs of the American Mathematical Society, Number 683, 2000.
[4] Burris & Sankappanavar: A Course in Universal Algebra, Springer-Verlag
1981.
[5] Albrecht Dold & Dieter Puppe: Homologie nicht-additiver Funktoren. Anwendungen, Annales de l’institut Fourier, tome 11, 1961.
[6] Eilenberg & Mac Lane: On the Groups H pΠ, nq, II — Methods of Computation,
Annals of Mathematics Volume 70, no. 1, 1954.
[7] Torsten Ekedahl: On minimal models in integral homotopy theory, Homotopy, Homology and Applications 4, no. 2, part 1, 2002.
[8] Jesse Elliott: Binomial rings, integer-valued polynomials, and λ-rings, Journal
of Pure and Applied Algebra 207, 2006.
[9] Peter Freyd: Abelian Categories: An Introduction to the Theory of Functors,
Harper & Row 1964.
[10] Eric M. Friedlander & Andrei Suslin: Cohomology of finite group schemes
over a field, Inventiones Mathematicae 127, 1997.
[11] Solomon W. Golomb: Iterated Binomial Coefficients, The American Mathematical Monthly, volume 87, no. 9, 1980.
[12] Philip Hall: The Edmonton Notes on Nilpotent Groups, Queen Mary College
Mathematics Notes 1976.
[13] Nathan Jacobson: Lie Algebras, Dover Publications 1962.
[14] Jean-Pierre Jouanolou: Théorèmes de Bertini et Applications, Birkhäuser 1983.
217
218
Bibliography
[15] Jean-Louis Loday & Bruno Vallette: Algebraic operads, book in preparation.
[16] I. G. Macdonald: Symmetric Functions and Hall Polynomials, Oxford University Press 1995.
[17] Pirashvili, T. I.: Polinomialnye funktory, Akademiya Nauk Gruzinskoı̆
SSR. Trudy Tbilisskogo Matematicheskogo Instituta im. A. M. Razmadze
91, 1988.
[18] N. Popescu: Abelian Categories with Applications to Rings and Modules, Academic Press 1973.
[19] Norbert Roby: Lois polynomes et lois formelles en théorie des modules, Annales
scientifiques de l’É.N.S., 3e série, tome 80, no 3 1963.
[20] Pelle Salomonsson: Contributions to the Theory of Operads, doctoral dissertation, Department of Mathematics, University of Stockholm 2003.
Contents
Acknowledgements
5
Introduction
7
Polynomielle Functorer på Module-Categorier — Sedo-Lärande Tankor öfver Algebran
0. Preliminaries
§1. Set Theory . . . . . . . .
§2. Module Theory . . . . . .
§3. Category Theory. . . . . .
§4. Semi-Abelian Category Theory
§5. Abelian Category Theory . .
§6. Commutative Algebra . . . .
11
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
17
17
17
18
19
22
25
1. Numerical Rings
§1. Numerical Rings. . . . . . . .
§2. Elementary Identities . . . . . .
§3. Torsion . . . . . . . . . . .
§4. Uniqueness . . . . . . . . . .
§5. Embedding in Q-Algebras . . . .
§6. Iterated Binomial CoeHcients . . .
§7. Homomorphisms . . . . . . .
§8. Free Numerical Rings . . . . . .
§9. Numerical Transfer. . . . . . .
§10. The Nilradical . . . . . . . .
§11. Numerical Ideals and Factor Rings .
§12. Finitely Generated Numerical Rings
§13. Modules . . . . . . . . . . .
§14. Exponentiation . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
27
28
31
31
33
33
34
34
35
36
37
38
39
41
42
.
.
.
.
.
.
2. Multi-Sets
45
§1. Multi-Sets . . . . . . . . . . . . . . . . . . . . 45
§2. Multations . . . . . . . . . . . . . . . . . . . . 48
§3. Confluent Products . . . . . . . . . . . . . . . . . 49
219
220
Contents
§4. The Multi-Set Category . . . . . . . . . . . . . . .
§5. Multi-Sets on Multi-Sets . . . . . . . . . . . . . . .
§6. Partitions and Compositions . . . . . . . . . . . . .
50
52
53
3. Mazes
§1. Mazes. . . . . . . . . . . .
§2. The Labyrinth Category . . . . .
§3. Operations on Mazes . . . . . .
§4. The Quotient Labyrinth Categories .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
55
. 56
. 57
. 60
. 62
4. Multi-Sets versus Mazes
§1. The Ariadne Functor . . . . .
§2. Pure Mazes . . . . . . . . .
§3. The Theseus Functor . . . . .
§4. The Category of Correspondences
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
65
. 66
. 69
. 70
. 73
5. Polynomial Maps
§1. Polynomiality . . . . . . .
§2. Polynomial Maps . . . . .
§3. Numerical Maps . . . . . .
§4. The Augmentation Algebras .
§5. Properties of Numerical Maps.
§6. Strict Polynomial Maps . . .
§7. The Divided Power Algebras .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
77
. 79
. 82
. 83
. 85
. 88
. 91
. 94
6. Polynomial Functors
§1. Module Functors . . . . . . . . . .
§2. Polynomial Functors . . . . . . . . .
§3. Numerical Functors . . . . . . . . .
§4. Properties of Numerical Functors . . . .
§5. The Hierarchy of Numerical Functors . . .
§6. Strict Polynomial Functors . . . . . . .
§7. The Hierarchy of Strict Polynomial Functors
§8. Homogeneous Functors . . . . . . . .
§9. Analytic Functors . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
97
98
100
100
102
105
108
110
111
112
7. Deviations and Cross-Effects
§1. The Deviations . . .
§2. The Cross-Evects . .
§3. The Multi-Deviations .
§4. The Multi-Cross-Evects
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
117
118
123
126
128
8. Projective Generators
§1. The Fundamental Module Functor . . . . . . . . . . .
§2. The Classical Yoneda Correspondence . . . . . . . . . .
§3. The Fundamental Numerical Functor . . . . . . . . . .
133
133
134
135
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
Contents
§4.
§5.
§6.
§7.
§8.
§9.
§10.
221
The Numerical Yoneda Correspondence . .
The Deviated Power Functors . . . . . .
The Labyrinthine Yoneda Correspondence .
The Fundamental Homogeneous Functor .
The Homogeneous Yoneda Correspondence.
The Divided Power Functors . . . . . .
The Multi-Set Yoneda Correspondence . .
9. Module Representations
§1. The Divided Power Map . . .
§2. Module Functors . . . . .
§3. Numerical Functors . . . .
§4. Homogeneous Functors . . .
§5. Quasi-Homogeneous Functors
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
136
137
141
144
145
147
150
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
153
154
158
160
161
162
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
163
164
170
171
174
175
178
179
184
184
11. Numerical versus Strict Polynomial Functors
§1. Quadratic Functors. . . . . . . . . .
§2. The Ariadne Thread . . . . . . . . .
§3. Numerical versus Strict Polynomial Functors
§4. Restriction and Extension of Scalars . . .
§5. Torsion-Free Functors. . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
187
189
191
193
195
197
12. Polynomial Monads and Operads
§1. Classical Operads . . . . . . . . .
§2. Schur Operads . . . . . . . . . .
§3. The Induction Product of Schur Operads.
§4. The Plethysm of Schur Operads . . . .
§5. Sur Operads . . . . . . . . . . .
§6. The Induction Product of Sur Operads .
§7. The Plethysm of Sur Operads . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
203
203
206
209
209
211
213
214
.
.
.
.
.
10. Combinatorial Representations
§1. Module Functors . . . . . .
§2. Polynomial Functors . . . . .
§3. Numerical Functors . . . . .
§4. Quasi-Homogeneous Functors .
§5. Quadratic Functors. . . . . .
§6. Labyrinth Modules . . . . . .
§7. Homogeneous Functors . . . .
§8. Homogeneous Quadratic Functors
§9. Schur Modules . . . . . . .
.
.
.
.
.
.
.